announced the judgment of the Court and delivered an opinion, in which The Chief Justice and Me. Justice Stewaet joined and in Parts I, II, III-A, III-B, III-C, and III-E of which Me. Justice Powell joined.
The Occupational Safety and Health Act of 1970 (Act), 84 Stat. 1590, 29 U. S. C. § 651 et seq., was enacted for the purpose of ensuring safe and healthful working conditions for every working man and woman in the Nation. This litigation concerns a standard promulgated by the Secretary of Labor to regulate occupational exposure to benzene, a substance which has been shown to cause cancer at high exposure levels. The principal question is whether such a showing is a sufficient basis for a standard that places the most stringent limitation on exposure to benzene that is technologically and economically possible.
The Act delegates broad authority to the Secretary to promulgate different kinds of standards. The basic definition *612of an “occupational safety and health standard” is found in § 3 (8), which provides:
“The term ‘occupational safety and health standard’ means a standard which requires conditions, or the adoption or use of one or more practices, means, methods, operations, or processes, reasonably necessary or appropriate to provide safe or healthful employment and places of employment.” 84 Stat. 1591, 29 U. S. C. § 652 (8).
Where toxic materials or harmful physical agents are concerned, a standard must also comply with § 6 (b)(5), which provides:
“The Secretary, in promulgating standards dealing with toxic materials or harmful physical agents under this subsection, shall set the standard which most adequately assures, to the .extent feasible, on the basis of the best available evidence, that no employee will suffer material impairment of health or functional capacity even if such employee has regular exposure to the hazard dealt with by such standard for the period of his working life. Development of standards under this subsection shall be based upon research, demonstrations, experiments, and such other information as may be appropriate. In addition to the attainment of the highest degree of health and safety protection for the employee, other considerations shall be the latest available scientific data in the field, the feasibility of the standards, and experience gained under this and other health and safety laws.” 84 Stat. 1594, 29 U. S. C. § 655 (b)(5).1
*613Wherever the toxic material to be regulated is a carcinogen, the Secretary has taken the position that no safe exposure level can be determined and that § 6 (b) (5) requires him to set an exposure limit at the lowest technologically feasible level that will not impair the viability of the industries regulated. In this case, after having determined that there is a causal connection between benzene and leukemia (a cancer of the white blood cells), the Secretary set an exposure limit on airborne concentrations of benzene of one part benzene per million parts of air (1 ppm), regulated dermal and eye contact with solutions containing benzene, and imposed complex monitoring and medical testing requirements on employers whose workplaces contain 0.5 ppm or more of benzene. 29 CFR §§ 1910.1028 (c), (e) (1979).
On pre-enforcement review pursuant to 29 U. S. C. § 655 (f), the United States Court of Appeals for the Fifth Circuit held the regulation invalid. American Petroleum Institute v. OSHA, 581 F. 2d 493 (1978). The court concluded that the Occupational Safety and Health Administration (QSHA) 2 had exceeded its standard-setting authority because it had not shown that the new benzene exposure limit was “reasonably necessary or appropriate to provide safe or healthful employment” as required by § 3 (8),3 and because § 6 (b)(5) *614does “not give OSHA the unbridled discretion to adopt standards designed to create absolutely risk-free workplaces regardless of costs.” 4 Reading the two provisions together, the Fifth Circuit held that the Secretary was under a duty to determine whether the benefits expected from the new standard bore a reasonable relationship to the costs that it imposed. Id., at 503. The court noted that OSHA had made an estimate of the costs of compliance, but that the record lacked substantial evidence of any discernible benefits.5
We agree with the’ Fifth Circuit's holding that § 3 (8) requires the Secretary to find, as a threshold matter, that the *615toxic substance in question poses a significant health risk in the workplace and that a new, lower standard is therefore “reasonably necessary or appropriate to provide safe or healthful employment and places of employment.” Unless^ and until such a finding is made, it is not necessary to address the further question whether the Court of Appeals correctly held that there must be a reasonable correlation between costs/ and benefits, or whether, as the federal parties argue, the Secretary is then required by §6 (b)(5) to promulgate a standard that goes as far as technologically and economically possible to eliminate the risk.
Because these are unusually important cases of first impression, we have reviewed the record with special care. In this opinion, we (1) describe the benzene standard, (2) analyze the Agency’s rationale for imposing a 1 ppm exposure limit, (3) discuss the controlling legal issues, and (4) comment briefly on the dermal contact limitation.
I
Benzene is a familiar and important commodity. It is a colorless, aromatic liquid that evaporates rapidly under ordinary atmospheric conditions. Approximately 11 billion pounds of benzene were produced in the United States in 1976. Ninety-four percent of that total was produced by the petroleum and petrochemical industries, with the remainder produced by the steel industry as a byproduct of coking operations. Benzene is used in manufacturing a variety of products including motor fuels (which may contain as much as 2% benzene), solvents, detergents, pesticides, and other organic chemicals. 43 Fed. Reg. 5918 (1978).
The entire population of the United States is exposed to small quantities of benzene, ranging from a few parts per billion to 0.5 ppm, in the ambient air. Tr. 1029-1032. Over one million workers are subject to additional low-level exposures as a consequence of their employment. The majority of these employees work in gasoline service stations, benzene *616production (petroleum refineries and coking operations), chemical processing, benzene transportation, rubber manufacturing, and laboratory operations.6
Benzene is a toxic substance. Although it could conceivably cause harm to a person who swallowed or touched it, the principal risk of harm comes from inhalation of benzene vapors. When these vapors are inhaled, the benzene diffuses through the lungs and is quickly absorbed into the blood. *617Exposure to high concentrations produces an almost immediate effect on the central nervous system. Inhalation of concentrations of 20,000 ppm can be fatal within minutes; exposures in the range of 250 to 500 ppm can cause vertigo, nausea, and other symptoms of mild poisoning. 43 Fed. Reg. 5921 (1978). Persistent exposures at levels above 25-40 ppm may lead to blood deficiencies and diseases of the blood-forming organs, including aplastic anemia, which is generally fatal.
Industrial health experts have long been aware that exposure to benzene may lead to various types of nonmalignant diseases. By 1948 the evidence connecting high levels of benzene to serious blood disorders had become so strong that the Commonwealth of Massachusetts imposed a 35 ppm limitation on workplaces within its jurisdiction. In 1969 the American National Standards Institute (ANSI) adopted a national consensus standard of 10 ppm averaged over an 8-hour period with a ceiling concentration of 25 ppm for 10-minute periods or a maximum peak concentration of 50 ppm. Id., at 5919. In 1971, after the Occupational Safety and Health Act was passed, the Secretary adopted this consensus standard as the federal standard, pursuant to 29 U. S. C. § 655 (a).7
*618As early as 1928, some health experts theorized that there might also be a connection between benzene in the workplace and leukemia.8 In the late 1960’s and early 1970’s a number of epidemiological studies were published indicating that workers exposed to high concentrations of benzene were subject to a significantly increased risk of leukemia.9 In a 1974 report recommending a permanent standard for benzene, the National Institute for Occupational Safety and Health *619(NIOSH), OSHA’s research arm,10 noted that these studies raised the “distinct possibility” that benzene caused leukemia. But, in light of the fact that all known cases had occurred at very high exposure levels, NIOSH declined to recommend a change in the 10 ppm standard, which it considered sufficient to protect against nonmalignant diseases. NIOSH suggested that further studies were necessary to determine conclusively whether there was a link between benzene and leukemia and, if so, what exposure levels were dangerous.11
Between 1974 and 1976 additional studies were published which tended to confirm the view that benzene can cause leukemia, at least when exposure levels are high.12 In an *620August 1976 revision of its earlier recommendation, NIOSH stated that these studies provided “conclusive” proof of a causal connection between benzene and leukemia. 1 Record, Ex. 2-5, p. 100. Although it acknowledged that none of the intervening studies had provided the dose-response data it had found lacking two years earlier, id., at 9, NIOSH nevertheless recommended that the exposure limit be set as low as possible. As a result of this recommendation, OSHA contracted with a consulting firm to do a study on the costs to industry of complying with the 10 ppm standard then in effect or, alternatively, with whatever standard would be the lowest feasible. Tr. 505-506.
In October 1976, NIOSH sent another memorandum to OSHA, seeking acceleration of the rulemaking process and “strongly” recommending the issuance of an emergency temporary standard pursuant to § 6 (c) of the Act, 29 U. S. C. § 655 (c),13 for benzene and two other chemicals believed to *621be carcinogens. NIOSH recommended that a 1 ppm exposure limit be imposed for benzene.14 1 Record, Ex. 2-6. Apparently because of the NIOSH recommendation, OSHA asked its consultant to determine the cost of complying with a 1 ppm standard instead of with the “minimum feasible” standard. Tr. 506-507. It also issued voluntary guidelines for benzene, recommending that exposure levels be limited to 1 ppm on an 8-hour time-weighted average basis wherever possible. 2 Record, Ex. 2-44.
In the spring of 1976, NIOSH had selected two Pliofilm plants in St. Marys and Akron, Ohio, for an epidemiological study of the link between leukemia and benzene exposure. In April 1977, NIOSH forwarded an interim report to OSHA indicating at least a fivefold increase in the expected incidence of leukemia for workers who had been exposed to ben*622zene at the two plants from 1940 to 1949.15 The report submitted to OSHA erroneously suggested that exposures in the two plants had generally been between zero and 15 ppm during the period in question.16 As a result of this new evidence *623and the continued prodding of NIOSH, 1 Record, Ex. 2-7, OSHA did issue an emergency standard, effective May 21. 1977, reducing the benzene exposure limit from 10 ppm to 1 ppm, the ceiling for exposures of up to 10 minutes from 25 ppm to 5 ppm, and eliminating the authority for peak concentrations of 50 ppm. 42 Fed. Reg. 22516 (1977). In its explanation accompanying the emergency standard, OSHA stated that benzene had been shown to cause leukemia at exposures below 25 ppm and that, in light of its consultant’s report, it was feasible to reduce the exposure limit to 1 ppm. Id., at 22517, 22521.
On May 19, 1977, the Court of Appeals for the Fifth Circuit entered a temporary restraining order preventing the emergency standard from taking effect. Thereafter, OSHA abandoned its efforts to make the emergency standard effective and instead issued a proposal for a permanent standard patterned almost entirely after the aborted emergency standard. Id., at 27452.
In its published statement giving notice of the proposed permanent standard, OSHA did not ask for comments as to whether or not benzene presented a significant health risk at exposures of 10 ppm or less. Rather, it asked for comments as to whether 1 ppm was the minimum feasible exposure limit.17 Ibid. As OSHA’s Deputy Director of Health Standards, Grover Wrenn, testified at the hearing, this formulation *624of the issue tó be considered by the Agency was consistent with OSHA’s general policy with respect to carcinogens.18 Whenever a carcinogen is involved, OSHA will presume that no safe level of exposure exists in the absence of clear proof establishing such a level and will accordingly set the exposure limit at the lowest level feasible.19 The proposed 1 ppm ex*625posure limit in this case thus was established not on the basis of a proven hazard at 10 ppm, but rather on the basis of “OSHA’s best judgment at the time of the proposal of the feasibility of compliance with the proposed standard by the [a]ffected industries.” Tr. 30. Given OSHA’s cancer policy, it was in fact irrelevant whether there was any evidence at all of a leukemia risk at 10 ppm. The important point was that there was no evidence that there was not some risk, however small, at that level. The fact that OSHA did not ask for comments on whether there was a safe level of exposure for benzene was indicative of its further view that a demonstration of such absolute safety simply could not be made.20
Public hearings were held on the proposed standard, commencing on July 19, 1977. The final standard was issued on February 10, 1978. 29 CFR § 1910.1028 (1979).21 In its final form, the benzene standard is designed to protect workers from whatever hazards are associated with low-level benzene *626exposures by requiring employers to monitor workplaces to determine the level of exposure, to provide medical examinations when the level rises above 0.5 ppm, and to institute whatever engineering or other controls are necessary to keep exposures at or below 1 ppm.
In the standard as originally proposed by OSHA, the employer’s duty to monitor, keep records, and provide medical examinations arose whenever any benzene was present in a workplace covered by the rule.22 Because benzene is omnipresent in small quantities, NIOSH and the President’s Council on- Wage and Price Stability recommended the use of an “action level” to trigger monitoring and medical examination requirements. Tr. 1030-1032; App. 121-133. OSHA accepted this recommendation, providing under the final standard that, if initial monitoring discloses benzene concentrations below 0.5 ppm averaged over an 8-hour work day, no further action is required unless there is a change in the company’s practices.23 If exposures are above the action *627level, but below the 1 ppm exposure limit, employers are required to monitor exposure levels on a quarterly basis and to provide semiannual medical examinations for their exposed employees. Neither the concept of an action level, nor the specific level selected by OSHA, is challenged in this proceeding.
Whenever initial monitoring indicates that employees are subject to airborne concentrations of benzene above 1 ppm averaged over an 8-hour workday, with a ceiling of 5 ppm for any 15-minute period, employers are required to modify their plants or institute work practice controls to reduce exposures within permissible limits. Consistent with OSHA’s general policy, the regulation does not allow respirators to be used if engineering modifications are technologically feasible.24 Employers in this category are also required to perform monthly monitoring so long as their workplaces remain above 1 ppm, provide semiannual medical examinations to exposed workers, post signs in and restrict access to “regulated areas” where the permissible exposure limit is exceeded, and conduct employee training programs where necessary.
The standard also places strict limits on exposure to liquid *628benzene. As originally framed, the standard totally prohibited any skin or eye contact with any liquid containing any benzene. Ultimately, after the standard was challenged, OSHA modified this prohibition by excluding liquids containing less than 0.5% benzene. After three years, that exclusion will be narrowed to liquids containing less than 0.1% benzene.
The permanent standard is expressly inapplicable to the storage, transportation, distribution, sale, or use of gasoline or other fuels subsequent to discharge from bulk terminals.25 This exception is particularly significant in light of the fact that over 795,000 gas station employees, who are exposed to an average of 102,700 gallons of gasoline (containing up to 2% benzene) annually, are thus excluded from the protection of the standard.26
As presently formulated, the benzene standard is an expensive way of providing some additional protection for a relatively small number of employees. According to OSHA’s figures, the standard will require capital investments in engineering controls of approximately $266 million, first-year operating costs (for monitoring, medical testing, employee training, and respirators) of $187 million to $205 million and *629recurring annual costs of approximately $34 million.27 43 Fed. Reg. 5934 (1978). The figures outlined in OSHA’s explanation of the costs of compliance to various industries indicate that only 35,000 employees would gain any benefit from the regulation in terms of a reduction in their exposure to benzene.28 Over two-thirds of these workers (24,450) are employed in the rubber-manufacturing industry. Compliance costs in that industry are estimated to be rather low with no capital costs and initial operating expenses estimated at only $34 million ($1,390 per employee); recurring annual costs would also be rather low, totaling less than $1 million. By contrast, the segment of the petroleum refining industry that produces benzene would be required to incur $24 million in capital costs and $600,000 in first-year operating expenses to provide additional protection for 300 workers .($82,000 per employee), while the petrochemical industry would be required to incur $20.9 million in capital costs and $1 million in initial operating expenses for the benefit of 552 employees ($39,675 per employee).29 Id., at 5936-5938.
*630Although OSHA did not quantify the benefits to each category of worker in terms of decreased exposure to benzene, it appears from the economic impact study done at OSHA’s direction that those benefits may be relatively small. Thus, although the current exposure limit is 10 ppm, the actual exposures outlined in that study are often considerably lower. For example, for the period 1970-1975 the petrochemical industry reported that, out of a total of 496 employees exposed to benzene, only 53 were exposed to levels between 1 and 5 ppm and only 7 (all at the same plant) were exposed to between 5 and 10 ppm. 1 Economic Impact Statement, p. 4-6, Table 4-2, 11 Record, Ex. 5A, p. 4-6, Table 4-2. See also id., Tables 4.3-4.8 (indicating sample exposure levels in various industries).
II
The critical issue at this point in the litigation is whether the Court of Appeals was correct in refusing to enforce the 1 ppm exposure limit on the ground that it was not supported by appropriate findings.30
*631Any discussion of the 1 ppm exposure limit must, of course, begin with the Agency’s rationale for imposing that limit.31 The written explanation of the standard fills 184 pages of the printed appendix. Much of it is devoted to a discussion of the voluminous evidence of the adverse effects of exposure to benzene at levels of concentration well above 10 ppm. This discussion demonstrates that there is ample justification for regulating occupational exposure to benzene and that the prior limit of 10 ppm, with a ceiling of 25 ppm (or a peak of 50 ppm) was reasonable. It does not, however, provide direct support for the Agency’s conclusion that the limit should be reduced from 10 ppm to 1 ppm.
The evidence in the administrative record of adverse effects of benzene exposure at 10 ppm is sketchy at best. OSHA noted that there was “no dispute” that certain nonmalignant blood disorders, evidenced by a reduction in the level of red or white cells or platelets in the blood, could result from exposures of 25-40 ppm. It then stated that several studies had indicated that relatively slight changes in normal blood values could result from exposures below 25 ppm and perhaps below 10 ppm. OSHA did not attempt to make any estimate based on these studies of how significant the risk of nonmalignant disease would be at exposures of 10 ppm or less.32 Rather, it stated that because of the lack of data concerning the linkage between low-level exposures and blood abnormalities, it was impossible to construct a dose-response *632curve at this time.33 OSHA did conclude, however, that the studies demonstrated that the current 10 ppm exposure limit was inadequate to ensure that no single worker would suffer a nonmalignant blood disorder as a result of benzene exposure. Noting that it is “customary” to set a permissible exposure limit by applying a safety factor of 10-100 to the lowest level at which adverse effects had been observed, the Agency stated that the evidence supported the conclusion that the limit should be set at a point “substantially less than 10 ppm” even if benzene’s leukemic effects were not considered. 43 Fed. Reg. 5924-5925 (1978). OSHA did not state, however, that the nonmalignant effects of benzene exposure justified a reduction in the permissible exposure limit to 1 ppm.34
OSHA also noted some studies indicating an increase in chromosomal aberrations in workers chronically exposed to *633concentrations of benzene “probably less than 25 ppm.”35 However, the Agency took no definitive position as to what these aberrations meant in terms of demonstrable health effects and stated that no quantitative dose-response relationship had yet been established. Under these circumstances, chromosomal effects were categorized by OSHA as an “adverse biological event of serious concern which may pose or reflect a potential health risk and as such, must be considered in the larger purview of adverse health effects associated with benzene. Id., at 5932-5934.
With respect to leukemia, evidence of an increased risk (i. e., a risk greater than that borne by the general population) due to benzene exposures at or below 10 ppm was even sketchier. Once OSHA acknowledged that the NIOSH study it had relied upon in promulgating the emergency standard did not support its earlier view that benzene had been shown to cause leukemia at concentrations below 25 ppm, see n. 12, supra, there was only one study that provided any evidence of such an increased risk. That study, conducted by the Dow Chemical Co., uncovered three leukemia deaths, versus 0.2 expected deaths, out of a population of 594 workers; it appeared that the three workers had never been exposed to more than 2 to 9 ppm of benzene. The authors of the study, however, concluded that it could not be viewed as proof of a relationship between low-level benzene exposure and leukemia because all three workers had probably been occupationally exposed to a number of other potentially carcinogenic chemicals at other points in their careers and because no leukemia deaths had been uncovered among workers who had been exposed to much higher levels of benzene. In its explanation of the permanent standard, OSHA stated that the possibility that these three leukemias had been caused by benzene exposure could not be *634ruled out and that the study, although not evidence of an increased risk of leukemia at 10 ppm, was therefore “consistent with the findings of many studies that there is an excess leukemia risk among benzene exposed employees.” 43 Fed. Reg. 5928 (1978). The Agency made no finding that the Dow study, any other empirical evidence, or any opinion testimony demonstrated that exposure to benzene at or below the 10 ppm level had ever in fact caused leukemia. See 581 F. 2d, at 503, where the Court of Appeals noted that OSHA was “unable to point to any empirical evidence documenting a leukemia risk at 10 ppm. . . .”
In the end OSHA’s rationale for lowering the permissible exposure limit to 1 ppm was based, not on any finding that leukemia has ever been caused by exposure to 10 ppm of benzene and that it will not be caused by exposure to 1 ppm, but rather on a series of assumptions indicating that some leuke-mias might result from exposure to 10 ppm and that the number of cases might be reduced by reducing the exposure level to 1 ppm. In reaching that result, the Agency first unequivocally concluded that benzene is a human carcinogen.36 Second, it concluded that industry had failed to prove that there is a safe threshold level of exposure to benzene below which no excess leukemia cases would occur. In reaching this conclusion OSHA rejected industry contentions that certain epidemiological studies indicating no excess risk of leukemia among workers exposed at levels below 10 ppm were sufficient to establish that the threshold level of safe exposure was at or above *63510 ppm.37 It also rejected an industry witness’ testimony that a dose-response curve could be constructed on the basis of the reported epidemiological studies and that this curve indicated that reducing the permissible exposure limit from 10 to 1 ppm would prevent at most one leukemia and one other cancer death every six years.38
Third, the Agency applied its standard policy with respect to carcinogens,39 concluding that, in the absence of definitive *636proof of a safe level, it must be assumed that any level above zero presents some increased risk of cancer.40 As the federal parties point out in their brief, there are a number of scientists and public health specialists who subscribe to this view, theorizing that a susceptible person may contract cancer from the absorption of even one molecule of a carcinogen like benzene. Brief for Federal Parties 18-19.41
*637Fourth, the Agency reiterated its view of the Act, stating that it was required by § 6 (b) (5) to set the standard either at the level that has been demonstrated to be safe or at the lowest level feasible, whichever is higher. If no safe level is established, as in this case, the Secretary’s interpretation of the statute automatically leads to the selection of an exposure limit that is the lowest feasible.42 Because of benzene’s importance to the economy, no one has ever suggested that it would be feasible to eliminate its use entirely, or to try to limit exposures to the small amounts that are omnipresent. Rather, the Agency selected 1 ppm as a workable exposure level, see n. 14, supra, and then determined that compliance with that level was technologically feasible and that “the economic impact of . . . [compliance] will not be such as to threaten the financial welfare of the affected firms or the general economy.” 43 Fed. Reg. 5939 (1978). It therefore held that 1 ppm was the minimum feasible exposure level within the meaning of § 6 (b)(5) of the Act.
Finally, although the Agency did not refer in its discussion of the pertinent legal authority to any duty to identify the anticipated benefits of the new standard, it did conclude that some benefits were likely to result from reducing the exposure limit from 10 ppm to 1 ppm. This conclusion was based, again, not on evidence, but rather on the assumption that the risk of leukemia will decrease as exposure levels decrease. Although the Agency had found it impossible to construct a dose-response curve that would predict with any accuracy the *638number of leukemias that could be expected to result from exposures at 10 ppm, at 1 ppm, or at any intermediate level, it nevertheless “determined that the benefits of the proposed standard are likely to be appreciable.” 43 43 Fed. Beg. 5941 (1978). In light of the Agency’s disavowal of any ability to determine the numbers of employees likely to be adversely affected by exposures of 10 ppm, the Court of Appeals held this finding to be unsupported by the record. 581 F. 2d, at 503.44
It is noteworthy that at no point in its lengthy explanation did the Agency quote or even cite § 3 (8) of the Act. It made no finding that any of the provisions of the new standard were “reasonably necessary or appropriate to provide safe or healthful employment and places of employment.” Nor did it allude to the possibility that any such finding might have been appropriate.
*639Ill
Our resolution of the issues in these cases turns, to a large extent, on the meaning of and the relationship between § 3 (8), which defines a health and safety standard as a standard that is “reasonably necessary and appropriate to provide safe or healthful employment/’ and § 6 (b)(5), which directs the Secretary in promulgating a health and safety standard for toxic materials to “set the standard which most adequately assures, to the extent feasible, on the basis of the best available evidence, that no employee will suffer material impairment of health or functional capacity. . . .”
In the Government’s view, § 3 (8)’s definition of the term “standard” has no legal significance or at best merely requires that a standard not be totally irrational. It takes the position that § 6 (b) (5) is controlling and that it requires OSHA to promulgate a standard that either gives an absolute assurance of safety for each and every worker or reduces exposures to the lowest level feasible. The Government interprets “feasible” as meaning technologically achievable at a cost that would not impair the viability of the industries subject to the regulation. The respondent industry representatives, on the other hand, argue that the Court of Appeals was correct in holding that the “reasonably necessary and appropriate” language of § 3 (8), along with the feasibility requirement of § 6 (b) (5), requires the Agency to quantify both the costs and the benefits of a proposed rule and to conclude that they are roughly commensurate.
In our view, it is not necessary to decide whether either the Government or industry is entirely correct. For we think it is clear that § 3 (8) does apply to all permanent standards promulgated under the Act and that it requires the Secretary, before issuing any standard, to determine that it is reasonably necessary and appropriate to remedy a significant risk of material health impairment. Only after the Secretary has made the threshold determination that such a risk exists *640with respect to a toxic substance, would it be necessary to decide whether § 6 (b)(5) requires him to select the most protective standard he can consistent with economic and technological feasibility, or whether, as respondents argue, the benefits of the regulation must be commensurate with the costs of its implementation. Because the Secretary did not make the required threshold finding in these cases, we have no occasion to determine whether costs must be weighed against benefits in an appropriate case
A
Under the Government’s view, §3 (8), if it has any substantive content at all,45 merely requires OSHA to issue stand*641ards that are reasonably calculated to produce a safer or more healthy work environment. Tr. of Oral Arg. 18, 20. Apart from this minimal requirement of rationality, the Government argues that § 3 (8) imposes no limits on the Agency’s power, and thus would not prevent it from requiring employers to do whatever would be “reasonably necessary” to eliminate all risks of any harm from their workplaces.46 With respect to toxic substances and harmful physical agents, the Government takes an even more extreme position. Relying on § 6 (b) (5)’s direction to set a standard “which most adequately assures... that no employee will suffer material impairment of health or functional capacity,” the Government contends that the Secretary is required to impose standards that either guarantee workplaces that are free from any risk of material health impairment, however small, or that come as close as possible to doing so without ruining entire industries.
If the purpose of the statute were to eliminate completely and with absolute certainty any risk of serious harm, we would agree that it would be proper for the Secretary to interpret §§ 3 (8) and 6 (b) (5) in this fashion. But we think it is clear that the statute was not designed to require employers to provide absolutely risk-free workplaces whenever it is technologically feasible to do so, so long as the cost is not great enough to destroy an entire industry. Rather, both the language and structure of the Act, as well as its legislative history, indicate that it was intended to require the elimination, as far as feasible, of significant risks of harm.
*642B
By empowering the Secretary to promulgate standards that are “reasonably necessary or appropriate to provide safe or healthful employment and places of employment,” the Act implies that, before promulgating any standard, the Secretary must make a finding that the workplaces in question are not safe. But “safe” is not the equivalent of “risk-free.” There are many activities that we engage in every day — such as driving a car or even breathing city air — that entail some risk of accident or material health impairment; nevertheless, few people would consider these activities “unsafe.” Similarly, a workplace can hardly be considered “unsafe” unless it threatens the workers with a significant risk of- harm.
Therefore, before he can promulgate any permanent health or safety standard, the Secretary is required to make a threshold finding that a place of employment is unsafe — in the sense that significant risks are present and can be eliminated or lessened by a change in practices. This requirement applies to permanent standards promulgated pursuant to § 6 (b)(5), as well as to other types of permanent standards. For there is no reason why § 3 (8)’s definition of a standard should not be deemed incorporated by reference into § 6 (b)(5). The standards promulgated pursuant to § 6 (b)(5) are just one species of the genus of standards governed by the basic requirement. That section repeatedly uses the term “standard” without suggesting any exception from, or qualification of, the general definition; on the contrary, it directs the Secretary to select “the standard” — that is to say, one of various possible alternatives that satisfy the basic definition in § 3 (8) — that is most protective.47 Moreover, requiring the *643Secretary to make a threshold finding of significant risk is consistent with the scope of the regulatory power granted to him by §6 (b)(5), which empowers the Secretary to promulgate standards, not for chemicals and physical agents generally, but for “toxic materials” and “harmful physical agents.” 48
This interpretation of §§ 3 (8) and 6 (b)(5) is supported by the other provisions of the Act. Thus, for example, § 6 (g) provides in part that
“[i]n determining the priority for establishing standards under this section, the Secretary shall give due regard to the urgency of the need for mandatory safety and health standards for particular industries, trades, *644crafts, occupations, businesses, workplaces or work environments.”
The Government has expressly acknowledged that this section requires the Secretary to undertake some cost-benefit analysis before he promulgates any standard, requiring the elimination of the most serious hazards first.49 If such an analysis must precede the promulgation of any standard, it seems manifest that Congress intended, at a bare minimum, that the Secretary find a significant risk of harm and therefore a probability of significant benefits before establishing a new standard.
Section 6 (b)(8) lends additional support to this analysis. That subsection requires that, when the Secretary substantially alters an existing consensus standard, he must explain how the new rule will “better effectuate” the purposes of the Act.50 If this requirement was intended to be more than a meaningless formality, it must be read to impose upon the Secretary the duty to find that an existing national consensus standard is not adequate to protect workers from a continuing and significant risk of harm. Thus, in this case, the Secretary was required to find that exposures at the current permissible *645exposure level of 10 ppm present a significant risk of harm in the workplace.
In the absence of a clear mandate in the Act, it is unreasonable to assume that Congress intended to give the Secretary the unprecedented power over American industry that would result from the Government’s view of §§ 3 (8) and 6 (b)(5), coupled with OSHA’s cancer policy. Expert testimony that a substance is probably a human carcinogen — either because it has caused cancer in animals or because individuals have contracted cancer following extremely high exposures — would justify the conclusion that the substance poses some risk of serious harm no matter how minute the exposure and no matter how many experts testified that they regarded the risk as insignificant. That conclusion would in turn justify pervasive regulation limited only by the constraint of feasibility. In light of the fact that there are literally thousands of substances used in the workplace that have been identified as carcinogens or suspect carcinogens, the Government’s theory would give OSHA power to impose enormous costs that might produce little, if any, discernible benefit.51
*646If the Government were correct in arguing that neither §3(8) nor §6 (b)(5) requires that the risk from a toxic substance be quantified sufficiently to enable the Secretary to characterize it as significant in an understandable way, the statute would make such a “sweeping delegation of legislative power” that it might be unconstitutional under the Court's reasoning in A. L. A. Schechter Poultry Corp. v. United States, 295 U. S. 495, 539, and Panama Refining Co. v. Ryan, 293 U. S. 388. A construction of the statute that avoids this kind of open-ended grant should certainly be favored.
C
The legislative history also supports the conclusion that Congress was concerned, not with absolute safety, but with the elimination of significant harm. The examples of industrial hazards referred to in the Committee hearings and debates all involved situations in which the risk was unquestionably significant. Tor example, the Senate Committee on Labor and Public Welfare noted that byssinosis, a disabling lung disease caused by breathing cotton dust, affected as many as 30% of the workers in carding or spinning rooms in some American cotton mills and that as many as 100,000 active or retired workers were then suffering from the disease. It also noted that statistics indicated that 20,000 out of 50,000 workers who had performed insulation work were likely to die of asbestosis, lung cancer, or mesothelyioma as a result of breathing asbestos fibers. Another example given of an occupational health hazard that would be controlled by the Act was betanaphthylamine, a “chemical so toxic that any exposure at all is likely to cause the development of bladder cancer over a period of years.” S. Eep. No. 91-1282, pp. 3-4 (1970); Legislative History of the Occupational Safety and Health Act of 1970 (Committee Print compiled for the Senate Committee on Labor and Public Welfare), pp. 143-144 (1971) (hereafter Leg. Hist.).
Moreover, Congress specifically amended § 6 (b) (5) to make *647it perfectly clear that it does not require the Secretary to promulgate standards that would assure an absolutely risk-free workplace. Section 6 (b) (5) of the initial Committee bill provided that
“[t]he Secretary, in promulgating standards under this subsection, shall set the standard which most adequately and feasibly assures, on the basis of the best available evidence, that no employee will suffer any impairment of health or functional capacity, or diminished life expectancy even if such employee has regular exposure to the hazard dealt with by such standard for the period of his working life.” (Emphasis supplied.) S. 2193, 91st Cong., 2d Sess., p. 39 (1970), Leg. Hist. 242.
On the floor of the Senate, Senator Dominick questioned the wisdom of this provision, stating:
“How in the world are we ever going to live up to that? What are we going to do about a place in Florida where mosquitoes are getting at the employee — perish the thought that there may be mosquitoes in Florida? But there are black flies in Minnesota and Wisconsin. Are we going to say that if employees get bitten by those for the rest of their lives they will not have been done any harm at all? Probably they will not be, but do we know?” 116 Cong. Rec. 36522 (1970), Leg. Hist. 345.
He then offered an amendment deleting the entire subsection.52 *648After discussions with the sponsors of the Committee bill, Senator Dominick revised his amendment. Instead of deleting the first sentence of § 6 (b) (5) entirely, his new amendment limited the application of that subsection to toxic materials and harmful physical agents and changed “any” impairment of health to “material” impairment.53 In discussing this change, Senator Dominick noted that the Committee’s bill read as if a standard had to “assure that, no matter what anybody was doing, the standard would protect him for the rest of his life against any foreseeable hazard.” Such an “unrealistic standard,” he stated, had not been intended by the sponsors of the bill. Rather, he explained that the intention of the bill, as implemented by the amendment, was to require the Secretary
“to use his best efforts to promulgate the best available standards, and in so doing, ... he should take into account that anyone working in toxic agents and physical *649agents which might be harmful may be subjected, to such conditions for the rest of his working life, so that we can get at something which might not be toxic now, if he works in it a short time, but if he works in it the rest of his life might be very dangerous; and we want to make sure that such things are taken into consideration in establishing standards.” 116 Cong. Rec., at 37622-37623, Leg. Hist. 502-503.54
Senator Williams, one of the sponsors of the Committee bill, agreed with the interpretation, and the amendment was adopted.
In their reply brief the federal parties argue that the Dominick amendment simply means that the Secretary is not required to eliminate threats of insignificant harm; they argue that § 6 (b) (5) still requires the Secretary to set standards that ensure that not even one employee will be subject to any risk of serious harm — no matter how small that risk may be.55 *650This interpretation is at odds with Congress’ express recognition of the futility of trying to make all workplaces totally risk-free. Moreover, not even OSHA follows this interpretation of § 6 (b) (5) to its logical conclusion. Thus, if OSHA is correct that the only no-risk level for leukemia due to benzene exposure is zero and if its interpretation of § 6 (b) (5) is correct, OSHA should have set the exposure limit as close to zero as feasible. But OSHA did not go about its task in that way. Rather, it began with a 1 ppm level, selected at least in part to ensure that employers would not be required to eliminate benzene concentrations that were little greater than the so-called “background” exposures experienced by the population at large. • See n. 14, supra. Then, despite suggestions by some labor unions that it was feasible for at least some industries to reduce exposures to well below 1 ppm,56 OSHA decided to apply the same limit to all, largely as a matter of administrative convenience. 43 Fed. Reg. 5947 (1978).
OSHA also deviated from its own interpretation of § 6 (b) (5) in adopting an action level of 0.5 ppm below which monitoring and medical examinations are not required. In light of OSHA’s cancer policy, it must have assumed that some employees would be at risk because of exposures below 0.5 ppm. These employees would thus presumably benefit from medical examinations, which might uncover any benzene-related problems. OSHA’s consultant advised the Agency that it was technologically and economically feasible to require that such examinations be provided. Nevertheless, OSHA adopted an action level, largely because the insignificant ben*651efits of giving such examinations and performing the necessary-monitoring did not justify the substantial cost.57
OSHA’s concessions to practicality in beginning with a 1 ppm exposure limit and using an action level concept implicitly adopt an interpretation of the statute as not requiring regulation of insignificant risks.58 It is entirely consistent with this interpretation to hold that the Act also requires the Agency to limit its endeavors in the standard-setting area to eliminating significant risks of harm.
Finally, with respect to the legislative history, it is important to note that Congress repeatedly expressed its concern about allowing the Secretary to have too much power over American industry. Thus, Congress refused to give the Secretary the power to shut down plants unilaterally because of an imminent danger, see Whirlpool Corp. v. Marshall, 445 U. S. 1, and narrowly circumscribed the Secretary’s power to issue temporary emergency standards.59 This effort by *652Congress to limit the Secretary’s power is not consistent with a view that the mere possibility that some employee somewhere in the country may confront some risk of cancer is a sufficient basis for the exercise of the Secretary’s power to require the expenditure of hundreds of millions of dollars to minimize that risk.
D
Given the conclusion that the Act empowers the Secretary to promulgate health and safety standards only where a significant risk of harm exists, the critical issue becomes how to define and allocate the burden of proving the significance of the risk in a case such as this, where scientific knowledge is imperfect and the precise quantification of risks is therefore impossible. The Agency’s position is that there is substantial evidence in the record to support its conclusion that there is no absolutely safe level for a carcinogen and that, therefore, the burden is properly on industry to prove, apparently beyond a shadow of a doubt, that there is a safe level for benzene exposure. The Agency argues that, because of the uncertainties in this area, any other approach would render it helpless, forcing it to wait for the leukemia deaths that it believes are likely to occur60 before taking any regulatory action.
*653We disagree. As we read the statute, the burden was on the Agency to show, on the basis of substantial evidence, that it is at least more likely than not that long-term exposure to 10 ppm of benzene presents a significant risk of material health impairment. Ordinarily, it is the proponent of a rule or order who has the burden of proof in administrative proceedings. See 5 U. S. C. § 556 (d). In some cases involving toxic substances, Congress has shifted the burden of proving that a particular substance is safe onto the party opposing the proposed rule.61 The fact that Congress did not follow this course in enacting the Occupational Safety and Health Act indicates that it intended the Agency to bear the normal burden of establishing the need for a proposed standard.
In this case OSHA did not even attempt to carry its burden of proof. The closest it came to making a finding that benzene presented a significant risk of harm in the workplace was its statement that the benefits to be derived from lowering the permissible exposure level from 10 to 1 ppm were "likely” to be “appreciable.” The Court of Appeals held that this finding was not supported by substantial evidence. Of greater importance, even if it were supported by substantial evidence, such a finding would not be sufficient to satisfy the Agency’s obligations under the Act.
The inadequacy of the Agency’s findings can perhaps be *654illustrated best by its rejection of industry testimony that a dose-response curve can be formulated on the basis of current ' epidemiological evidence and that, even under the most conservative extrapolation theory, current exposure levels would cause at most two deaths out of a population of about 30,000 workers every six years. See n. 38, supra. In rejecting this testimony, OSHA made the following statement:
“In the face of the record evidence of numerous actual deaths attributable to benzene-induced leukemia and other fatal blood diseases, OSHA is unwilling to rely on the hypothesis that at most two cancers every six years would be prevented by the proposed standard. By way of example, the Infante study disclosed seven excess leukemia deaths in a population of about 600 people over a 25-year period. While the Infante study involved higher exposures then those currently encountered, the incidence rates found by Infante, together with the numerous other cases reported in the literature of benzene leukemia and other fatal blood diseases, make it difficult for OSHA to rely on the [witness’] hypothesis to assure the statutorily mandated protection of employees. In any event, due to the fact that there is no safe level of exposure to benzene and that it is impossible to precisely quantify the anticipated benefits, OSHA must select the level of exposure which is most protective of exposed employees.” 43 Fed. Reg. 5941 (1978).
There are three possible interpretations of OSHA’s stated reason for rejecting the witness’ testimony: (1) OSHA considered it probable that a greater number of lives would be saved by lowering the standard from 10 ppm; (2) OSHA thought that saving two lives every six years in a work force of 30,000 persons is a significant savings that makes it reasonable and appropriate to adopt a new standard; or (3) even if the small number is not significant and even if the savings may be even smaller, the Agency nevertheless believed it had *655a statutory duty to select the level of exposure that is most protective of the exposed employees if it is economically and technologically feasible to do so. Even if the Secretary did not intend to rely entirely on this third theory, his construction of the statute would make it proper for him to do so. Moreover, he made no express findings of fact that would support his 1 ppm standard on any less drastic theory. Under these circumstances, we can hardly agree with the Government that OSHA discharged its duty under the Act.
Contrary to the Government’s contentions, imposing a burden on the Agency of demonstrating a significant risk of harm will not strip it of its ability to regulate carcinogens, nor will it require the Agency to wait for deaths to occur before taking any action. First, the requirement that a “significant” risk be identified is not a mathematical straitjacket. It is the Agency’s responsibility to determine, in the first instance, what it considers to be a “significant” risk. Some risks are plainly acceptable and others are plainly unacceptable. If, for example, the odds are one in a billion that a person will die from cancer by taking a drink of chlorinated water, the risk clearly could not be considered significant. On the other hand, if the odds are one in a thousand that regular inhalation of gasoline vapors that are 2% benzene will be fatal, a reasonable person might well consider the risk significant and take appropriate steps to decrease or eliminate it. Although the Agency has no duty to calculate the exact probability of harm, it does have an obligation to find that a significant risk is present before it can characterize a place of employment as “unsafe.” 62
*656Second, OSHA is not required to support its finding that a significant risk exists with anything approaching scientific certainty. Although the Agency’s findings must be supported by substantial evidence, 29 U. S. C. § 655 (f), § 6 (b)(5) specifically allows the Secretary to regulate on the basis of the “best available evidence.” As several Courts of Appeals have held, this provision requires a reviewing court to give OSHA some leeway where its findings must be made on the frontiers of scientific knowledge. See Industrial Union Dept., AFL-CIO v. Hodgson, 162 U. S. App. D. C. 331, 340, 499 F. 2d 467, 476 (1974); Society of the Plastics Industry, Inc. v. OSHA, 509 F. 2d 1301, 1308 (CA2 1975), cert. denied, 421 U. S. 992. Thus, so long as they are supported by a body of reputable scientific thought, the Agency is free to use conservative assumptions in interpreting the data with respect to carcinogens, risking error on the side of overprotection rather than underprotection.63
Finally, the record in this case and OSHA’s own rulings on other carcinogens indicate that there axe a number of ways in which the Agency can make a rational judgment about the *657relative significance of the risks associated with exposure to a particular carcinogen.64
It should also be noted that, in setting a permissible exposure level in reliance on less-than-perfect methods, OSHA would have the benefit of a backstop in the form of monitor*658ing and medical testing. Thus, if OSHA properly determined that the permissible exposure limit should be set at 5 ppm, it could still require monitoring and medical testing for employees exposed to lower levels.65 By doing so, it could keep a constant check on the validity of the assumptions made in developing the permissible exposure limit, giving it a sound evidentiary basis for decreasing the limit if it was initially set too high.66 Moreover, in this way it could ensure that workers who were unusally susceptible to benzene could be removed from exposure before they had suffered any permanent damage.67
E
Because our review of these cases has involved a more detailed examination of the record than is customary, it must *659be emphasized that we have neither made any factual determinations of our own, nor have we rejected any factual findings made by the Secretary. We express no opinion on what factual findings this record might support, either on the basis of empirical evidence or on the basis of expert testimony; nor do we express any opinion on the more difficult question of what factual determinations would warrant a conclusion that significant risks are present which make promulgation of a new standard reasonably necessary or appropriate. The standard must, of course, be supported by the findings actually made by the Secretary, not merely by findings that we believe he might have made.
In this case the record makes it perfectly clear that the Secretary relied squarely on a special policy for carcinogens that imposed the burden on industry of proving the existence of a safe level of exposure, thereby avoiding the Secretary’s threshold responsibility of establishing the need for more stringent standards. In so interpreting his statutory authority, the Secretary exceeded his power.
IV
Throughout the administrative proceedings, the dermal contact issue received relatively little attention. In its proposed rule OSHA recommended a total ban on skin and eye contact with liquid benzene on the basis of its policy that “in dealing with a carcinogen, all potential routes of exposure (i. e., inhalation, ingestion, and skin absorption) [should] be limited to the extent feasible.” 43 Fed. Reg. 5948 (1978). There was little opposition to this requirement at the hearing on the proposed rule, apparently because the proposed rule also excluded from both the permissible exposure level and the dermal contact ban work operations involving liquid mixtures containing 1% (and after one year, 0.1%) or less benzene.
In its final standard, however, OSHA eliminated the percentage exclusion for liquid benzene, on the ground that there was no predictable correlation between the percentage of ben*660zene in a liquid and the airborne exposure arising from it. See n. 22, supra. Although the extent to which liquid benzene is absorbed through the skin is concededly unknown, OSHA also refused to exempt any liquids, no matter how little benzene they contained, from -the ban on dermal contact. In support of this position it stated that there was no evidence to “suggest that the absorption rate depends on the amount of benzene present in the liquid.” 43 Fed. Reg. 5948-5949 (1978).
After the permanent standard was promulgated, OSHA received a number of requests from various industries that the percentage exclusion for liquids containing small amounts of benzene be reinstated. Those concerned with airborne exposures argued that they should not be required to monitor workplaces simply because they handled petroleum-based products in which benzene is an unavoidable contaminant. Others concerned with the dermal contact ban made similar arguments. In particular, tire manufacturers argued that it was impossible for them to- comply with the ban because gloves cannot be worn during certain tire-building operations in which solvents are used and solvents containing absolutely no benzene are not commercially available.
Because of these requests, OSHA held a new series of hearings and promulgated an amendment to the rule, reinstating the percentage exclusion, but lowering it from the proposed 1% to 0.5%. The Agency did, however, provide for a 3-year grace period before the exclusion dropped to 0.1%, rather than the one year that had originally been proposed. In explaining its amendment, OSHA reiterated its policy with respect to carcinogens, stating that, because there is no absolutely safe level for any type of exposure, exposures by whatever route must be limited to the extent feasible. For airborne exposures, a zero permissible exposure limit had not been feasible. However, in most industries a ban on any dermal contact was feasible since compliance could be achieved simply by the use of protective clothing, such as impermeable *661gloves. The Agency recognized that the dermal contact ban could present a problem for tire manufacturers, but stated that the percentage exclusion would alleviate the problem, because solvents containing 0.5% or less benzene were available in sufficient quantities. Although it noted that solvents containing 0.1% or less benzene were not then available in quantity, the Agency stated that a 3-year grace period would be sufficient to “allow time for increased production of solvents containing lower amounts of benzene and for development and evaluation of alternative methods of compliance with the standard’s dermal provision.” Id., at 27968-27969.
The Court of Appeals struck down the dermal contact prohibition on two grounds. First, it held that the record did not support a finding that the ban would result in quantifiable benefits in terms of a reduced leukemia risk; therefore, it was not “reasonably necessary” within the meaning of § 3 (8) of the Act. Second, the court held that the Agency’s conclusion that benzene may be absorbed through the skin was not based on the best available evidence as required by § 6 (b)(5). 581 F. 2d, at 505-506. On the second ground, the court noted that the evidence on the issue of absorption of benzene through the skin was equivocal, with some studies indicating that it could be absorbed and some indicating that it could not. All of these studies were relatively old and the only expert who had testified on the issue stated that a simple test was now available to determine, with a great deal of accuracy, whether and to what extent absorption will result. In light of 16(b)(5), which requires the Agency to promulgate standards on the basis of the “best available evidence” and “the latest available scientific data in the field,” the court held that where there is uncontradicted testimony that a simple test will resolve the issue, the Agency is required to acquire that information before “promulgating regulations which would require, an established industry to change long-followed work processes that are not demonstrably unsafe.” 581 F. 2d, at 508.
*662While the court below may have been correct in holding that, under the peculiar circumstances of this case, OSHA was required to obtain more information, there is no need for us to reach that issue. For, in order to justify a ban on dermal contact, the Agency must find that such a ban is “reasonably necessary and appropriate” to remove a significant risk of harm from such contact. The Agency did not make such a finding, but rather acted on the basis of the absolute, no-risk policy that it applies to carcinogens. Indeed, on this issue the Agency’s position is even more untenable, inasmuch as it was required to assume not only that benzene in small doses is a carcinogen, but also that it can be absorbed through the skin in sufficient amounts to present a carcinogenic risk. These assumptions are not a proper substitute for the findings of a significant risk of harm required by the Act.
The judgment of the Court of Appeals remanding the petition for review to the Secretary for further proceedings is affirmed.
It is so ordered.
The second and third sentences of this section, which impose feasibility limits on the Secretary and allow him to take into account the best available evidence in developing standards, may apply to all health and safety standards. This conclusion follows if the term “subsection” used in the second sentence refers to the entire subsection 6 (b) (which sets out procedures for the adoption of all types of health and safety standards), rather than simply to the toxic materials subsection, §6 (b)(5). While Me. Justice Marshall, post, at 694, and respondents agree with this *613position, see Brief for Respondents American Petroleum Institute et al. 39; see also Currie, OSHA, 1976 Am. Bar Found. Research J. 1107, 1137, n. 151, the Government does not, see Brief for Federal Parties 58; see also Berger & Riskin, Economic and Technological Feasibility in Regulating Toxic Substances Under the Occupational Safety and Health Act, 7 Ecology L. Q. 285, 294 (1978). There is no need for us to decide this issue in these cases.
OSHA is the administrative agency within the Department of Labor that is responsible for promulgating and enforcing standards under the Act. In this opinion, we refer to the “Secretary,” “OSHA” and the “Agency” interchangeably.
“The Act imposes on OSHA the obligation to enact only standards that are reasonably necessary or appropriate to provide safe or healthful workplaces. If a standard does not fit in this definition, it is not one that OSHA is authorized to enact.” 581 F. 2d, at 502.
“Although 29 U. S. C. A. § 655 (b) (5) requires the goal of attaining the highest degree of health and safety protection for the employee, it does not give OSHA the unbridled discretion to adopt standards designed to create absolutely risk-free workplaces regardless of Cost. To the contrary, that section requires standards to be feasible, and it contains a number of pragmatic limitations in the form of specific kinds of information OSHA must consider in enacting standards dealing with toxic materials. Those include 'the best available evidence,’ 'research, demonstrations, experiments, and such other information as may be appropriate,’ ‘the latest available scientific data in the field,’ and 'experience gained under this and other health and safety laws.’ Moreover, in standards dealing with toxic materials, just as with all other occupational safety and health standards, the conditions and other requirements imposed by the standard must be ‘reasonably necessary or appropriate to provide safe or healthful employment and places of employment.’ 29 U. S. C. A. § 652 (8).” Ibid.
“The lack of substantial evidence of discemable benefits is highlighted when one considers that OSHA is unable to point to any empirical evidence documenting a leukemia risk at 10 ppm even though that has been the permissible exposure limit since 1971. OSHA’s assertion that benefits from reducing the permissible exposure limit from 10 ppm to 1 ppm are likely to be appreciable, an assumption based only on inferences drawn from studies involving much higher exposure levels rather than on studies involving these levels or sound statistical projections from the high-level studies, does not satisfy the reasonably necessary requirement limiting OSHA’s action. Aqua Slide requires OSHA to estimate the extent of expected benefits in order to determine whether those benefits bear a reasonable relationship to the standard’s demonstrably high costs.” Id., at 503-504.
OSHA’s figures indicate that 795,000 service station employees have some heightened exposure to benzene as a result of their employment. See 2 U. S. Dept, of Labor, OSHA, Technology Assessment and Economic Impact Study of an OSHA Regulation for Benzene, p. D-7 (May 1977) (hereinafter Economic Impact Statement), 11 Record, Ex. 5B, p. D-7. These employees are specifically excluded from the regulation at issue in this case. See infra, at 628. OSHA states that another 629,000 employees, who are covered by the regulation, work in the other industries described. 43 Fed. Reg. 5935 (1978).
It is not clear from the record or its explanation of the permanent standard how OSHA arrived at the estimate of 629,000 exposed employees. OSHA’s consultant, Arthur D. Little, Inc., estimated that there were 191,000 exposed employees, 30,000 of whom were exposed to 1 ppm or more of benzene. 1 Economic Impact Statement, p. 3-5, 11 Record, Ex. 5A, p. 3-5. In its explanation of the permanent standard OSHA stated that there were 1,440 exposed employees who worked in benzene plants, 98,000 in other petroleum refineries, 24,000 in coke ovens, 4,000 in light oil plants, 2,760 in the petrochemical industry, 52,345 who worked in bulk terminals, 23,471 drivers who loaded benzene from those terminals, 74,000 in oil and gas production, 17,000 in pipeline work, 100 at tank-car facilities, 200 at tank-truck facilities, 480 on barges, 11,400 in tire-manufacturing plants, and 13,050 in other types of rubber production. 43 Fed. Reg. 5936-5938 (1978). Although OSHA gave no estimate for laboratory workers, the A. D. Little study indicated that there were 25,000 exposed workers in that industry. These figures add up to 347,246 exposed employees — approximately 282,000 less than the overall estimate of 629,000. It is possible that some or all of these employees work in the “other industries” briefly described in OSHA’s explanation; these are primarily small firms that manufacture adhesives, paint and ink or that use benzene solvents Id., at 5939. No estimate of the number of exposed employees in those industries or the aggregate cost of compliance by those industries is given either by OSHA or by A D, Little in its consulting report.
Section 6 (a) of the Act, as set forth in 29 U. S. C. § 655 (a), provides:
“Without regard to chapter 5 of Title 5 or to the other subsections of this section, the Secretary shall, as soon as practicable during the period beginning with the effective date of this chapter and ending two years after such date, by rule promulgate as an occupational safety or health standard any national consensus standard, and any established Federal standard, unless he determines that the promulgation of such a standard would not result in improved safety or health for specifically designated employees. In the event of conflict among any such standards, the Secretary shall promulgate the standard which assures the greatest protection of the safety or health of the affected employees.”
In this case the Secretary complied with the directive to choose the most protective standard by selecting the ANSI standard of 10 ppm, rather than the 25 ppm standard adopted by the American Conference of Government Industrial Hygienists. 43 Fed. Reg. 5919 (1978).
See Delore & Borgomano, Leucémie aigué au cours de l’intoxication benzenique. Sur l’origine toxique de certaines leucémies aigues et leurs relations avec.Ies anémies graves, 9 Journal de Medécine de Lyon 227 (1928). A translation of that document appears in the benzene administrative record. 2 Record, Ex. 2-60. See also Hunter, Chronic Exposure to Benzene (Benzol). II. The Clinical Effects, 21 J. Ind. Hyg. & Toxicol. 331 (1939), 3 Record, Ex. 2-74, which refers to “leucemia” as a side effect of chronic exposure to benzene.
Dr. Muzaffer Aksoy, a Turkish physician who testified at the hearing on the proposed benzene standard, did a number of studies concerning the effects of benzene exposure on Turkish shoemakers. The workers in Dr. Aksoy’s studies used solvents containing large percentages of benzene and were constantly exposed to high concentrations of benzene vapors (between 150 and 650 ppm) under poorly ventilated and generally unhygienic conditions. See Aksoy, Acute Leukemia Due to Chronic Exposure to Benzene, 52 Am. J. of Medicine 160 (1972), 1 Record, Ex. 2-29; Aksoy, Benzene (Benzol): Its Toxicity and Effects on the Hematopoietic System, Istanbul Faculty of Medicine Monograph Series No. 51 (1970), 2 Record, Ex. 2-55; Aksoy, Erdem, & DinCol, Leukemia in Shoe-Workers Exposed Chronically to Benzene, 44 Blood 837 (1974), 2 Record, Ex. 2-53 (reporting on 26 shoeworkers who had contracted leukemia from 1967 to 1973; this represented an incidence of 13 per 100,000 rather than the 6 cases per 100,000 that would normally be expected).
Dr. Enrico Vigliani also reported an excess number of leukemia cases among Italian shoemakers exposed to glues containing a high percentage of benzene and workers in rotogravure plants who had been exposed over long periods of time to inks and solvents containing as much as 60% benzene. See Vigliani & Saita, Benzene and Leukemia, 271 New Eng. J. of Medicine 872-876 (1964), 1 Record, Ex. 2-27; Forni & Vigliani, Chemical Leukemogenesis in Man, 7 Ser. Haemat. 211 (1974), 2 Record, Ex. 2-50.
Title 29 U. S. C. § 669 (a) (3) requires the Department of Health, Education, and Welfare (HEW) (now in part the Department of Health and Human Services) to develop “criteria” dealing with toxic materials and harmful physical agents that describe “exposure levels that are safe for various periods of employment.” HEW’s obligations under this section have been delegated to NIOSH, 29 U. S. C. § 671.
See Dept, of HEW, NIOSH, Criteria for a Recommended Standard— Occupational Exposure to Benzene 74-75 (Pub. No. 74-137, 1974), 1 Record, Ex. 2-3. In response to a letter from the Director of the Office of Standards Division, NIOSH stated that its 10 ppm standard was designed to protect against leukemia, as well as other health risks. NIOSH noted, however, that further research was necessary in order to establish adequate dose-response data for benzene and leukemia. 12 Record, Ex. 32A, 32B.
Aksoy published another study in 1976 reporting on an additional eight leukemia cases uncovered after 1973. In that article, he also noted that a 1969 ban on the use of benzene as a solvent had led to a decline in the number of reported leukemia cases beginning in 1974. Aksoy, Types of Leukemia in Chronic Benzene Poisoning, 55 Acta Haematologica 65 (1976), 1 Record, Ex. 2-30. Vigliani also noted a decline in leukemia cases in Italy after benzene was no longer used in glues and inks. See Vigliani & Forni, Benzene and Leukemia, 11 Environmental Res. 122 (1976), 1 Record, Ex. 2-15; Vigliani, Leukemia Associated with Benzene Exposure, 271 Annals N. Y. Acad, of Sciences 143 (1976), 2 Record, Ex. 2-49. In the latter study Vigliani noted that in the past 100% pure ben*620zene solvents had been used and workers had been exposed on a prolonged basis to concentrations of 200-500 ppm, with peaks of- up to 1500 ppm.
A number of epidemiological studies were also done among American rubber workers during this period. Dr. A. J. McMichael’s studies indicated a ninefold increase in the risk of contracting leukemia among workers who were heavily exposed in the 1940’s and 1950’s to pure benzene used as a solvent. McMichael, Spirtas, Kupper, & Gamble, Solvent Exposure and Leukemia Among Rubber Workers: An Epidemiologic Study, 17 J. of Occup. Med. 234, 238 (1975), 2 Record, Ex. 2-37. See also Andjel-kovic, Taulbee, & Symons, Mortality Experience of a Cohort of Rubber Workers, 1964-1973, 18 J. of Occup. Med. 387 (1976), 2 Record, Ex. 2-54 (also indicating an excess mortality rate from leukemia among rubber workers).
Section 655 (c) provides:
“(1) The Secretary shall provide, without regard to the requirements of chapter 5 of title 5, for an emergency temporary standard to take immediate effect upon publication in the Federal Register if he determines (A) that employees are exposed to grave danger from exposure to substances or agents determined to be toxic or physically harmful or from *621new hazards, and (B) that such emergency standard is necessary to protect employees from such danger.
“(2) Such standard shall be effective until superseded by a standard promulgated in accordance with the procedures prescribed in paragraph (3) of this subsection.
“(3) Upon publication of such standard in the Federal Register the Secretary shall commence a proceeding in accordance with subsection (b) of this section, and the standard as published shall also serve as a proposed rule for the proceeding. The Secretary shall promulgate a standard under this paragraph no later than six months after publication of the emergency standard as provided in paragraph (2) of this subsection.”
At the hearing on the permanent standard NIOSH representatives testified that they had selected 1 ppm initially in connection with the issuance of a proposed standard for vinyl chloride. In that proceeding they had discovered that 1 ppm was approximately the lowest level detectable through the use of relatively unsophisticated monitoring instruments. With respect to benzene, they also thought that 1 ppm was an appropriate standard because any lower standard might require the elimination of the small amounts of benzene (in some places up to 0.5 ppm) that are normally present in the atmosphere. Tr. 1142-1143. NIOSH’s recommendation was not based on any evaluation of the feasibility, either technological or economic, of eliminating all exposures above 1 ppm. Id., at 1156.
Seven fatalities from leukemia were discovered out of the 748 workers surveyed. However, Dr. Infante, who conducted the study, stated that his statistical techniques had probably underestimated the number of leukemia cases that had actually occurred. Id., at 747. The normal expected incidence of leukemia in such a population would be 1.4. 2 Record, Ex. 2-51, p. 6.
The authors’ statement with respect to exposure levels was based on a 1946 report by the Ohio Industrial Commission indicating that, after some new ventilation equipment had been installed, exposures at the St. Marys plant had been brought within “safe” limits, in most instances ranging from zero to 10 to 15 ppm. Id., at 3. As the authors later admitted, the level considered “safe” in 1946 was 100 ppm. Tr. 814-815. Moreover, only one of the seven workers who died of leukemia had begun working at St. Marys after 1946. Five of the others had worked at the Akron plant, which employed 310 of the 748 workers surveyed. Id., at 2537-2538. A 1948 report by the Ohio Department of Health indicated exposure levels at the Akron plant of well over 100 ppm, with excursions in some areas up tc 1,000 ppm. 17 Record, Ex. 84A, App. A, pp. 61-62. Surveys taken in the intervening years, as well as testimony by St. Marys employees at the hearing on the proposed standard, Tr. 3432-3437, indicated that both of the plants may have had relatively high exposures through the 1970’s.
Industry representatives argued at the hearing that this evidence indicated that the exposure levels had been very high, as they had been in the other epidemiological studies conducted in the past. See Post-Hearing Brief for American Petroleum Institute in No. H-059 (OSHRC), pp. 23-37, 31 Record, Ex. 217-33, pp. 23-37. NIOSH witnesses, however, simply stated that actual exposure levels for the years in question could not be determined; they did agree, however, that their study should not be taken as proof of a fivefold increase in leukemia risk at 10-15 ppm. Tr. 814-815. In its explanation of the permanent standard, OSHA agreed with the NIOSH witnesses that no dose-response relationship could be inferred from the study:
“Comments at the hearing demonstrated that there were area exposures during this study period exceeding these levels [10-15 ppm], at times reaching values of hundreds of parts per million. Since no personal moni*623toring data are available, any conclusion regarding the actual individual time-weighted average exposure is speculative. Because of the lack of definitive exposure data, OSHA cannot derive any conclusions linking the excess leukemia risk observed with any specific exposure level.” 43 Fed. Reg. 5927 (1978).
OSHA also sought public comment as to whether certain industries should be exempt from compliance, whether the proposed compliance procedures and labeling techniques were adequate, what the environmental and economic consequences of the regulation would be, and whether it was feasible to replace benzene in solvents and other products of which it constituted more than 1%.
It became clear at the hearing that OSHA had not promulgated the proposed standard in response to any new concern about the nonmalignant effects of low-level benzene exposure. See Tr. 126-127:
“Is it accurate to say that the reason why the — why OSHA has proposed to reduce the exposure limits in the standard below the current levels is because of a perceived risk of leukemia, and not because of any new evidence it'has received that the current standards are inadequate to protect against acute or chronic benzene toxicity, other than leukemia?
“MR. WRENN: I think I will simply refer the part of my statement you were referring to, in which it says, it is however benzene’s leukemo-genicity which is of greatest concern to OSHA. That is certainly the central issue within the ETS [emergency temporary standard] and the proposed standard.”
Mr. Wrenn testified:
“The proposed standard requires that employee exposure to benzene in air be reduced to one part per million, with a five part per million ceiling allowable over any fifteen minute period during an eight hour work shift, and prohibits eye or prolonged skin contact with liquid benzene.
“This airborne exposure limit is based on OSHA’s established regulatory policy, that in the absence of a demonstrated safe level, or a no effect level for a carcinogen, it will be assumed that none exist, and that the agency will attempt to limit employee exposure to the lowest level feasible.” Id., at 29-30.
See also:
“MR. WARREN: Mr. Wrenn, in promulgating the emergency temporary, and proposed permanent, benzene standards, OSHA relies heavily, and I am quoting from your testimony now, on the regulatory policy that there is no safe level for carcinogens at any — for any exposed population, and the fact that leukemia, and a leukemogen is a carcinogen, is that correct?
“MR. WRENN: I believe that I stated that slightly differently in my oral summary of the statement than it is stated in the statement itself. I said that in the absence of a known or demonstrated safe level or no *625effect level, our policy is to assume that none exists, and to regulate accordingly.” Id., at 48-49.
“MR. WRENN: I would prefer to state it as I have on a couple of occasions already this morning, and that in the absence of a demonstrated safe level of exposure, we will assume that none exists for the purpose of regulatory policy.” Id., at 50.
In answer to the question of what demonstration would suffice to establish a “safe level,” Mr. Wrenn stated:
“I would like to draw a distinction, however, between what I have referred to as the demonstration that a safe level exists, arid speculation or elaborate theories that one may make, and I think that the agency in its history and very likely its future regulatory policy, would, in the face of evidence demonstrating that a carcinogenic hazard does exist or did exist, in this particular set of circumstances, would be very reluctant to accept as the basis for its regulatory decisions, a theoretical argument that a safe level may, in fact, exist for a particular substance.” Id., at 51-52.
A NIOSH representative who testified later put it more succinctly, stating that “. . . if benzene causes leukemia, and if leukemia is a cancer, then exposure really is almost moot.” Id., at 1007.
An amendment to the standard was promulgated on June 27, 1978. 43 Fed. Reg. 27962. See n. 22, infra.
Apart from its exclusion of gasoline storage and distribution facilities (an exclusion retained in the final rule, see text, at n. 25, infra), the proposed rule also excluded from coverage work operations in which liquid mixtures containing 1% or less benzene were used. After a year this exclusion was to be narrowed to operations where 0.1% benzene solutions were used. The rationale for the exclusion was that airborne exposures from such liquids would generally be within the 1 ppm limit. However, testimony at the hearing on the proposed rule indicated that there was no “consistent predictable relationship” between benzene content in a liquid and the resulting airborne exposure. Therefore, OSHA abandoned the idea of a percentage exclusion for liquid benzene in its final standard. 43 Fed. Reg. 5942 (1978).
OSHA later reconsidered its position and, in an amendment to the permanent standard, reinstated an exclusion for liquids, setting the level at 0.5%, to be reduced to 0.1% after three years, id., at 27962.
The exemption from the monitoring and medical testing portions of the standard for workplaces with benzene exposure levels below 0.5 ppm was not predicated on any finding that regulation of such workplaces was not feasible. OSHA’s consultant, Arthur D. Little, Inc., concluded that 1 ppm was a feasible exposure limit even assuming that there was no *627action level (or, to put it another way, assuming that the action level was zero). Rather, it was, as NIOSH witnesses stated, a practical decision based on a determination that, where benzene exposures are below 0.5 ppm, they will be unlikely ever to rise above the permissible exposure level of 1 ppm. NIOSH was also concerned that, in the absence of an action level, employers who used sophisticated analytical equipment might be required to monitor and provide medical examinations simply because of the presence of benzene in the ambient air. Tr. 1030-1032, 1133-1134.
Indeed, in its explanation of the standard OSHA states that an employer is required to institute engineering controls (for example, installing new ventilation hoods) even if those controls are insufficient, by themselves, to achieve compliance and respirators must therefore be used as well. 43 Fed. Reg. 5952 (1978). OSHA’s preference for engineering modifications is based on its opinion that respirators are rarely used properly (because they are uncomfortable, are often not properly fitted, etc.) and therefore cannot be considered adequate protective measures.
It is also inapplicable to work operations involving 0.5% liquid benzene (0.1% after three years), see n. 22, supra, and to the handling of benzene in sealed containers or systems, except insofar as employers are required to provide cautionary notices and appropriate employee training.
Prior to the introduction of the action-level concept, A. D. Little estimated that compliance costs for the service station industry might be as high as $4 billion. Tr. 508-509. Moreover, A. D. Little’s Economic Impact Statement indicated that service station employees were generally exposed to very low levels of benzene. 1 Economic Impact Statement, p. 4-21, 11 Record, Ex. 5A, p. 4-21. Still, in its explanation accompanying the permanent standard OSHA did not rule out regulation of this industry entirely, stating that it was in the process of studying whether and to what extent it should regulate exposures to gasoline in general. 43 Fed. Reg. 5943 (1978).
OSHA’s estimate of recurring annual costs was based on the assumption that the exposure levels it had projected would be confirmed by initial monitoring and that, after the first year, engineering controls would be successful in bringing most exposures within the 1 ppm limit. Under these circumstances, the need for monitoring, medical examinations, and respirators would, of course, be drastically reduced.
Three hundred of these employees work in benzene plants, 5,000 in other petroleum refineries, 4,000 in light oil plants, 552 in the petrochemical industry, 156 in benzene transportation, 1,250 in laboratories, 11,400 in tire-manufacturing plants, and 13,050 in other rubber-manufacturing plants. OSHA also estimated that another 16,216 workers (5,000 in petroleum refineries, 1,104 in the petrochemical industry, 7,300 in bulk terminals, 312 in benzene transportation, and 2,500 in laboratories) would be exposed to 0.5 to 1 ppm of benzene and thus would receive a benefit in terms of more comprehensive medical examinations. Id., at 5936-5938.
The high cost per employee in the latter two industries is attributable to OSHA’s policy of requiring engineering controls rather than allowing respirators to be used to reduce exposures to the permissible limit. The *630relatively low estimated cost per employee in the rubber industry is based on OSHA’s assumption that other solvents and adhesives can be substituted for those that contain benzene and that capital costs will therefore not be required.
The other issue before us is whether the Court of Appeals correctly refused to enforce the dermal contact ban. That issue is discussed in Part IV, infra.
In the court below respondents also challenged the monitoring and medical testing requirements, arguing that certain industries should have been totally exempt from them and that, as to other industries, the Agency had not demonstrated that all the requirements were reasonably necessary to ensure worker health and safety. They also argued that OSHA’s requirement that the permissible exposure limit be met through engineering controls rather than through respirators was not reasonably necessary under the Act. Because it invalidated the 1 ppm exposure limit, the Fifth Circuit had no occasion to deal with these issues, and they are not now before this Court.
As we have often held, the validity of an agency’s determination must be judged on the basis of the agency’s stated reasons for making that determination. See SEC v. Chenery Corp., 318 U. S. 80, 95 (“[A]n administrative order cannot be upheld unless the grounds upon which the agency acted in exercising its powers were those upon which its action can be sustained”); FPC v. Texaco Inc., 417 U. S. 380, 397; FTC v. Sperry & Hutchinson Co., 405 U. S. 233, 249.
As OSHA itself noted, some blood abnormalities caused by benzene exposure may not have any discernible health effects, while others may lead to significant impairment and even death. 43 Fed. Reg. 5921 (1978).
“A dose-response curve shows the relationship between different exposure levels and the risk of cancer [or any other disease] associated with those exposure levels. Generally, exposure to higher levels carries with it a higher risk, and exposure to lower levels is accompanied by a reduced risk.” 581 F. 2d, at 504, n. 24.
OSHA’s comments with respect to the insufficiency of the data were addressed primarily to the lack of data at low exposure levels. OSHA did not discuss whether it was possible to make a rough estimate, based on the more complete epidemiological and animal studies done at higher exposure levels, of the significance of the risks attributable to those levels, nor did it discuss whether it was possible to extrapolate from such estimates to derive a risk estimate for low-level exposures.
OSHA did not invoke the automatic rule of reducing exposures to the lowest limit feasible that it applies to cancer risks. Instead, the Secretary reasoned that prudent health policy merely required that the permissible exposure limit be set “sufficiently below the levels at which adverse effects have been observed to assure adequate protection for all exposed employees.” 43 Fed. Reg. 5925 (1978). While OSHA concluded that application of this rule would lead to an exposure limit “substantially less than 10 ppm,” it did not state either what exposure level it considered to present a significant risk of harm or what safety factor should be applied to that level to establish a permissible exposure limit.
While citing these studies, OSHA also noted that other studies of similarly exposed workers had not indicated any increased level of chromosome damage.
“The evidence in the record conclusively establishes that benzene is a human carcinogen. The determination of benzene’s leukemogenicity is derived from the evaluation of all the evidence in totality and is not based on any one particular study. OSHA recognizes, as indicated above that individual reports vary considerably in quality, and that some investigations have significant methodological deficiencies. While recognizing the strengths and weaknesses in individual studies, OSHA nevertheless concludes that the benzene record as a whole clearly establishes a causal relationship between benzene and leukemia.” Id., at 5931.
In rejecting these studies, OSHA stated that: “Although the epidemiological method can provide strong evidence of a causal relationship between exposure and disease in the case of positive findings, it is by its very nature relativelytcrude and an insensitive measure.” After noting a number of specific ways in which such studies are often defective, the Agency stated that it is “OSHA’s policy when evaluating negative studies, to hold them to higher standards of methodological accuracy." Id., at 5931-5932. Viewing the industry studies in this light, OSHA concluded that each of them had sufficient methodological defects to make them unreliable indicators of the safety of low-level exposures to benzene.
OSHA rejected this testimony in part because it believed the exposure data in the epidemiological studies to be inadequate to formulate a dose-response curve. It also indicated that even if the testimony was accepted— indeed as long as there was any increase in the risk of cancer — the Agency was under an obligation to “select the level of exposure which is most protective of exposed employees.” Id., at 5941.
In his dissenting opinion, Me. Justice Marshall states that the Agency did not rely “blindly on some Draconian carcinogen ‘policy’ ” in setting a permissible exposure limit for benzene. He points to the large number of witnesses the Agency heard and the voluminous record it compiled as evidence that it relied instead on the particular facts concerning benzene. With all due respect, we disagree with Me. Justice MaRshall’s interpretation of the Agency’s rationale for its decision. After hearing the evidence, the Agency relied on the same policy view it had stated at the outset, see supra, at 623-625, namely, that, in the absence of clear evidence to the contrary, it must be assumed that no safe level exists for exposure to a carcinogen. The Agency also reached the entirely predictable conclusion that industry had not carried its concededly impossible burden, see n. 41, infra, of proving that a safe level of exposure exists for benzene. As the Agency made clear later in its proposed generic cancer policy, see n. 51, infra, it felt compelled to allow industry witnesses to go over the same ground in each regulation dealing with a carcinogen, despite *636its policy view. The generic policy, which has not yet gone into effect, was specifically designed to eliminate this duplication of effort in each case by foreclosing industry from arguing that there is a safe level for the particular carcinogen being regulated. 42 Fed. Reg. 54154r-54155 (1977).
“As stated above, the positive studies on benzene demonstrate the causal relationship of benzene to the induction of leukemia. Although these studies, for the most part involve high exposure levels, it is OSHA’s view that once the carcinogenicity of a substance .has been established qualitatively, any exposure must be considered to be attended by risk when considering any given population. OSHA therefore believes that occupational exposure to benzene at low levels poses a carcinogenic risk to workers.” 43 Fed. Reg. 5932 (1978).
The so-called “one hit” theory is based on laboratory studies indicating that one molecule of a carcinogen may react in the test tube with one molecule of DNA to produce a mutation. The theory is that, if this occurred in the human body, the mutated molecule could replicate over a period of years and eventually develop into a cancerous tumor. See OSHA’s Proposed Rule on the Identification, Classification and Regulation of Toxic Substances Posing a Potential Carcinogenic Risk, 42 Fed. Reg. 54148, 54165-54167 (1977). Industry witnesses challenged this theory, arguing that the presence of several different defense mechanisms in the human body make it unlikely that a person would actually contract cancer as a result of absorbing one carcinogenic molecule. Thus, the molecule might be detoxified before reaching a critical site, damage to a DNA molecule might be repaired, or a mutated DNA molecule might be destroyed by the body’s immunological defenses before it could develop into a cancer. Tr. 2836.
In light of the improbability of a person’s contracting cancer as a result of a single hit, a number of the scientists testifying on both sides of the issue agreed that every individual probably does have a threshold exposure limit below which he or she will not contract cancer. See, e. g., id., at 1179-1181. The problem, however, is that individual susceptibility appears to vary greatly and there is at present no way to calculate each *637and every person’s threshold. Thus, even industry witnesses agreed that if the standard must ensure with absolute certainty that every single worker is protected from any risk of leukemia, only a zero exposure limit would suffice. Id., at 2492, 2830.
“There is no doubt that benzene is a carcinogen and must, for the protection and safety of workers, be regulated as such. Given the inability to demonstrate a threshold or establish a safe level, it is appropriate that OSHA prescribe that the permissible exposure to benzene be reduced to the lowest level feasible.” 43 Fed. Reg. 5932 (1978).
At an earlier point in its explanation, OSHA stated:
“There is general agreement that benzene exposure causes leukemia as well as other fatal diseases of the bloodforming organs. In spite of the certainty of this conclusion, there does not exist an adequate scientific basis for establishing the quantitative dose response relationship between exposure to benzene and the induction of leukemia and other blood diseases. The uncertainty in both the actual magnitude of expected deaths and in the theory of extrapolation from existing data to the OSHA exposure levels places the estimation of benefits on 'the frontiers of scientific knowledge.’ While the actual estimation of the number of cancers to be prevented is highly uncertain, the evidence indicates that the number may be appreciable. There is general agreement that even in the absence of the ability to establish a ‘threshold’ or ‘safe’ level for benzene and other carcinogens, a dose response relationship is likely to exist; that is, exposure to higher doses carries with it a higher risk of cancer, and conversely, exposure to lower levels is accompanied by a reduced risk, even though a precise quantitative relationship cannot be established.” Id., at 5940.
The court did, however, hold that the Agency’s other conclusions— that there is some risk of leukemia at 10 ppm and that the risk would decrease by decreasing the exposure limit to 1 ppm — were supported by substantial evidence. 581 F. 2d, at 503.
We cannot accept the argument that § 3 (8) is totally meaningless. The Act authorizes the Secretary to promulgate three different kinds of standards — national consensus standards, permanent standards, and temporary emergency standards. The only substantive criteria given for two of these — national consensus standards and permanent standards for safety hazards not covered by § 6 (b) (5) — are set forth in § 3. While it is true that § 3 is entitled “definitions,” that fact does not drain each definition of substantive content. For otherwise there would be no purpose in defining the critical terms of the statute. Moreover, if the definitions were ignored, there would be no statutory criteria at all to guide the Secretary in promulgating either national consensus standards or permanent standards other than those dealing with toxic materials and harmful physical agents. We may not expect Congress to display perfect craftsmanship, but it is unrealistic to assume that it intended to give no direction whatsoever to the Secretary in promulgating most of his standards.
The structure of the separate subsection describing emergency temporary standards, 29 U. S C. § 655 (c), quoted in n. 13, supra, supports this conclusion. It authorizes the Secretary to bypass the normal procedures for setting permanent standards if he makes two findings: (A) that employees are exposed to “grave danger” from exposure to toxic substances and (B) that an emergency standard is “necessary” to protect the employees from that danger Those findings are to be compared with those that are implicitly required by the definition of the permanent standard — (A) that there be a significant — as opposed to a “grave” — risk, and (B) that additional regulation is “reasonably necessary or appropriate” — as opposed to “necessary.” It would be anomalous for Congress to require specific find-*641mgs for temporary standards but to give the Secretary a carte blanche for permanent standards.
The Government does not concede that the feasibility requirement in the second sentence of § 6 (b) (5) applies to health and safety standards other than toxic substances standards. See n. 1, supra. However, even if it did, the Government’s interpretation of the term “feasible,” when coupled with its view of § 3 (8), would still allow the Agency to require the elimination of even insignificant risks at great cost, so long as an entire industry’s viability would not be jeopardized.
Section 6 (b) (5) parallels § 6 (a) in this respect. Section 6 (a) requires the Secretary, when faced with a choice between two national consensus standards, to choose the more protective standard, see n. 7, supra. Just as § 6 (a) does not suggest that this more protective standard need not meet the definition of a national consensus standard set forth in § 3 (9), *643so § 6 (b) (5) does not suggest that the most protective toxic material standard need not conform to the definition of a "standard” in § 3 (8).
The rest of §6 (b)(5), while requiring the Secretary to promulgate the standard that “most adequately assures . . . that no employee will suffer material impairment of health or functional capacity,” also contains phrases implying that the Secretary should consider differences in degrees of significance rather than simply a total elimination of all risks. Thus, the standard to be selected is one that “most adequately assures, to the extent feasible, on the basis of the best available evidence,” that no such harm will result. The Secretary is also directed to take into account “research, demonstrations, experiments, and such other information as may be appropriate” and to consider “[i]n addition to the attainment of the highest degree of health and safety protection for the employee ... the latest available scientific data in the field, the feasibility of the standards, and experience gained under this and other health and safety laws.”
Mr. Justice Marshall states that our view of § 3 (8) would make the first sentence in § 6 (b) (5) superfluous. We disagree. The first sentence of § 6 (b) (5) requires the Secretary to select a highly protective standard once he has determined that a standard should be promulgated. The threshold finding that there is a need for such a standard in the sense that there is a significant risk in the workplace is not unlike the threshold finding that a chemical is toxic or a physical agent is harmful. Once the Secretary has made the requisite threshold finding, § 6 (b) (5) directs him to choose the most protective standard that still meets the definition of a standard under §3 (8), consistent with feasibility.
“First, 29 U. S. C. § 655 (g) requires the Secretary to establish priorities in setting occupational health and safety standards so that the more serious hazards are addressed first. In setting such priorities the Secretary must, of course, consider the relative costs, benefits and risks.” Reply Brief for Federal Parties 13. The Government argues that the Secretary’s setting of priorities under this section is not subject to judicial review. Tr. of Oral Arg. 23. While we agree that a court cannot tell the Secretary which of two admittedly significant risks he should act to regulate first, this section, along with §§ 3 (8) and 6 (b) (5), indicates that the Act does limit the Secretary’s power to requiring the elimination of significant risks.
Section 6 (b) (8), as set forth in 29 U. S. C. § 655 (b) (8), provides:
“Whenever a rule promulgated by the Secretary differs substantially from an existing national consensus standard, the Secretary shall, at the same time, publish in the Federal Register a statement of the reasons why the rule as adopted will better effectuate the purposes of this chapter than the national consensus standard.”
OSHA’s proposed generic cancer policy, 42 Fed. Reg. 54148 (1977), indicates that this possibility is not merely hypothetical. Under its proposal, whenever there is a certain quantum of proof — either from animal experiments, or, less frequently, from epidemiological studies — that a substance causes cancer at any exposure level, an emergency temporary standard would be promulgated immediately, requiring employers to provide monitoring and medical examinations and to reduce exposures to the lowest feasible level. A proposed rule would then be issued along the same lines, with objecting employers effectively foreclosed from presenting evidence that there is little or no risk associated with current exposure levels. Id., at 54154-54155 ; 29 CFR, Part 1990 (1977).
The scope of the proposed regulation is indicated by the fact that NIOSH has published a list of 2,415 potential occupational carcinogens, NIOSH, Suspected Carcinogens: A Subfile of the Registry of Toxic Effects of Chemical Substances (HEW Pub. No. 77-149, 2d ed. 1976). OSHA has tentatively concluded that 269 of these substances have been proved to be carcinogens and therefore should be subject to full regulation. See OSHA Press Release, USDL 78-625 (July 14, 1978).
In criticizing the Committee bill, Senator Dominick also made the following observations:
“It is unrealistic to attempt, as this section apparently does, to establish a utopia free from any hazards. Absolute safety is an impossibility and it will only create confusion in the administration of this act for the Congress to set clearly unattainable goals.” 116 Cong. Rec. 37614 (1970), Leg. Hist. 480.
“But I ask, Mr. President, just thinking about that language, let us take a fellow who is a streetcar conductor or a bus conductor at the present time. How in the world, in the process of the pollution we have in the streets *648or in the process of the automobile accidents that we have all during a working day of any one driving a bus or trolley car, or whatever it may be, can we set standards that will make sure he will not have any risk to his life for the rest of his life? It is totally impossible for this to be put in a bill; and yet it is in the committee bill.” 116 Cong. Rec., at 37337, Leg. Hist. 423.
As an opponent of the legislation, Senator Dominick may have exaggerated the significance of the problem since the language in § 3 (8) already was sufficient to prevent the Secretary from trying “to establish a utopia free from any hazards.” Nevertheless, the fact that Congress amended the bill to allay Senator Dominick’s concern demonstrates that it did not intend the statute to achieve “clearly unattainable goals.”
Senator Dominick had also been concerned that the placement of the word “feasibly” could be read to require the Secretary to “ban all occupations in which there remains some risk of injury, impaired health, or life expectancy,” since the way to most “adequately” and “feasibly” assure absolute protection might well be to prohibit the occupation entirely. 116 Cong. Rec., at 36530, Leg. Hist. 366-367, In his final amendment, he attempted to cure this problem by relocating the feasibility requirement, changing “the standard which most adequately and feasibly assures” to “the standard which most adequately assures, to the extent feasible.”
Mr. Justice Marshall argues that Congress could not have thought § 3 (8) had any substantive meaning inasmuch as § 6 (b) (5), as originally drafted, applied to all standards and not simply to standards for toxic materials and harmful physical substances. However, as this legislative history indicates, it appears that the omission of the words “toxic substances” and “harmful physical agents” from the original draft of § 6 (b) (5) was entirely inadvertent. As Senator Dominick noted, the Committee had always intended that subsection to apply only to that limited category of substances. The reason that Congress drafted a special section for these substances was not, as Mr. Justice Marshall suggests, because it thought that there was a need for special protection in these areas. Rather, it was because Congress recognized that there were special problems in regulating health risks as opposed to safety risks. In the latter case, the risks are generally immediate and obvious, while in the former, the risks may not be evident until a worker has been exposed for long periods of time to particular substances. It was to ensure that the Secretary took account of these long-term risks that Congress enacted § 6 (b) (5).
Reply Brief for Federal Parties 24-26. While it is true that some of Senator Dominick’s comments were concerned with the relative unimportance of minor injuries (see his “fly” example quoted supra, at 647), it is *650clear that he was also concerned with the remote possibility of major injuries, see n. 52, supra.
One union suggested a 0.5 ppm permissible exposure limit for oil refineries and a 1 ppm ceiling (rather than a time-weighted average) exposure for all other industries, with no use of an action level, Tr. 1250, 1257. Another wanted a 1 ppm ceiling limit for all industries, id., at 3375-3376.
“A need for an action level is also suggested by the record evidence that some minimal exposure to benzene occurs naturally from animal and plant matter (Tr. 749-750; 750-760). Naturally occurring benzene concentrations, it appears, may range from 0.02 to 15 parts per billion (Ex. 117, p. 1). Additionally, it was suggested by certain employers that their operations be exempted from the requirements of the standard because these operations involve only intermittent and low level exposures to benzene. The use of the action level concept should accommodate these concerns in all cases where exposures are indeed extremely low since it substantially reduces the monitoring of employees who are below the action level and removes for these employees the requirements for medical surveillance. At the same time, employees with significant overexposure are afforded the full protection of the standard.” (Emphasis added.) 43 Fed. Reg. 5942 (1978).
The Government also states that it is OSHA’s policy to attempt to quantify benefits wherever possible. While this is certainly a reasonable position, it is not consistent with OSHA’s own view of its duty under § 6 (b)(5). In light of the inconsistencies in OSHA’s position and the legislative history of the Act, we decline to defer to the Agency’s interpretation.
In Florida Peach Growers Assn., Inc. v. U. S. Dept. of Labor, 489 F. 2d 120, 130, and n. 16 (CA5 1974), the court noted that Congress intended *652to restrict the use of emergency standards, which are promulgated without any notice or hearing. It held that, in promulgating an emergency standard, OSHA must find not only a danger of exposure or even some danger from exposure, but also a grave danger from exposure necessitating emergency action. Accord, Dry Color Mfrs. Assn., Inc. v. U. S. Dept. of Labor, 486 F. 2d 98, 100 (CA3 1973) (an emergency standard must be supported by something more than a possibility that a substance may cause cancer in man).
Congress also carefully circumscribed the Secretary’s enforcement powers by creating a new, independent board to handle appeals from citations issued by the Secretary for noncompliance with health and safety standards. See 29 U. S. C. §§ 659-661.
As noted above, OSHA acknowledged that there was no empirical evidence to support the conclusion that there was any risk whatsoever of deaths due to exposures at 10 ppm. What OSHA relied upon was a theory *653that, because leukemia deaths had occurred at much higher exposures, some (although fewer) were also likely to occur at relatively low exposures. The Court of Appeals specifically held that its conclusion that the number was “likely” to be appreciable was unsupported by the record. See supra, at 638.
See Environmental Defense Fund, Inc. v. EPA, 179 U. S. App. D. C. 43, 49, 57-63, 548 F. 2d 998, 1004, 1012-1018 (1977), cert. denied, 431 U. S. 925, where the court rejected the argument that the EPA has the-burden of proving that a pesticide is unsafe in order to suspend its registration under the Federal Insecticide, Fungicide, and Rodenticide Act. The court noted that Congress had deliberately shifted the ordinary burden of proof under the Administrative Procedure Act, requiring manufacturers to establish the continued safety of their products.
In his dissenting opinion, post, at 706, Mr. Justice Marshall states: “[W]hen the question involves determination of the acceptable level of risk, the ultimate decision must necessarily be based on considerations of policy as well as empirically verifiable facts. Factual determinations can at most define the risk in some statistical way; the judgment whether that risk is tolerable cannot be based solely on a resolution of *656the facts.” We agree. Thus, while the Agency must support its finding that a certain level of risk exists by substantial evidence, we recognize that its determination that a particular level of risk is "significant” will be based largely on policy considerations. At this point we have no need to reach the issue of what level of scrutiny a reviewing court should apply to the latter type of determination.
Mr. Justice Marshall states that, under our approach, the Agency must either wait for deaths to occur or must “deceive the public” by making a basically meaningless determination of significance based on totally inadequate evidence. Mr. Justice Marshall’s view, however, rests on the erroneous premise that the only reason OSHA did not attempt to quantify benefits in this case was because it could not do so in any reasonable manner. As the discussion of the Agency’s rejection of an industry attempt at formulating a dose-response curve demonstrates, however, see supra, at 653-655, the Agency’s rejection of methods such as dose-response curves was based at least in part on its view that nothing less than absolute safety would suffice.
For example, in the coke-oven emissions standard, OSHA had calculated that 21,000 exposed coke-oven workers had an annual excess mortality of over 200 and that the proposed standard might well eliminate the risk entirely. 41 Fed. Reg. 46742,46750 (1976), upheld in American Iron & Steel Inst. v. OSHA, 577 F. 2d 825 (CA3 1978), cert. granted, post, p. 909. In hearings on the coke-oven emissions standard, the Council on Wage and Price Stability estimated that 8 to 35 lives would be saved each year, out of an estimated population of 14,000 workers, as a result of the proposed standard. Although noting that the range of benefits would vary depending on the assumptions used, OSHA did not make a finding as to whether its own staff estimate or CWPS’s was correct, on the ground that it was not required to quantify the expected benefits of the standard or to weigh those benefits against the projected costs.
In other proceedings, the Agency has had a good deal of data from animal experiments on which it could base a conclusion on the significance of the risk. For example, the record on the vinyl chloride standard indicated that a significant number of animals had developed tumors of the liver, lung, and skin when they were exposed to 50 ppm of vinyl chloride over a period of 11 months. One hundred out of 200 animals died during that period. 39 Fed. Reg. 35890, 35891 (1974). Similarly, in a 1974 standard regulating 14 carcinogens, OSHA found that one of the substances had caused lung cancer in mice or rats at 1 ppm and even 0.1 ppm, while another had caused tumors in 80% of the animals subjected to high doses. Id., at 3756, 3757, upheld in Synthetic Organic Chemical Mfrs. Assn. v. Brennan, 503 F. 2d 1155 (CA3 1974), cert. denied, 420 U. S. 973, and Synthetic Organic Chemical Mfrs. Assn. v. Brennan, 506 F. 2d 385 (CA3 1974), cert. denied, 423 U. S. 830.
In this case the Agency did not have the benefit of animal studies, because scientists have been unable as yet to induce leukemia in experimental animals as a result of benzene- exposure.- It did, however, have a fair amount of epidemiological evidence, including both positive and negative studies. Although the Agency stated that this evidence was insufficient to construct a precise correlation between exposure levels and cancer risks, it would at least be helpful in determining whether it is more likely than not that there is a significant risk at 10 ppm.
See GAF Corp. v. Occupational Safety and Health Review Comm’n, 183 U. S. App. D. C. 20, 561 F. 2d 913 (1977), where the court upheld the asbestos standard insofar as it required employers to provide medical examinations for employees exposed to any asbestos fibers, even if they were exposed to concentrations below the permissible exposure limit.
The respondent industry representatives have never disputed OSHA’s power to require monitoring and medical examinations in general, although they did object to some of the specific requirements imposed in this case. See n. 30, supra. Because of our disposition of the case, we have no occasion to pass on these specific objections or to determine what cost-benefit considerations, if any, should govern the Agency’s imposition of such requirements.
This is precisely the type of information-gathering function that Congress had in mind when it enacted § 6 (b) (7), which empowers the Secretary to require medical examinations to be furnished to employees exposed to certain hazards and potential hazards “in order to most effectively determine whether the health of such employees is adversely affected by such exposure.” See S. Rep. No. 91-1282, p. 7 (1970), Leg. Hist. 147.
In its explanation of the final standard OSHA noted that there was some testimony that blood abnormalities would disappear after exposure had ceased. 43 Fed. Reg. 5946 (1978). Again, however, OSHA refused to rely on the hypothesis that this would always occur. Yet, in requiring medical examinations of employees exposed to between 0.5 ppm and 1 ppm, OSHA was essentially providing itself with the same kind of backstop.