Opinion for the Court filed by
Circuit Judge MIKVA. Concurring statement filed by Circuit Judge WILLIAMS. MIKVA, Circuit Judge:Three sets of petitioners seek review of a final rule of the Environmental Protection Agency (EPA or the agency) that promulgated recommended maximum contaminant levels (recommended levels) for eight volatile organic compounds (VOCs). The rule established recommended levels of zero for five VOCs that the EPA found to be known or probable carcinogens, a recommended level above zero for one compound the agency found to be a possible carcinogen, and recommended levels above' zero for two compounds the agency determined to be non-carcinogens. Several petitioners challenge the EPA’s general determination to set recommended levels for known or probable carcinogens at zero. In addition, two industrial petitioners challenge the inclusion of a particular VOC in the category of known or probable carcinogens, while petitioner Natural Resources Defense Council (NRDC) contests EPA’s decision to set a recommended level above zero for the compound determined to be a possible carcinogen. Finding the agency’s determinations to be well within the bounds of its authority under the Safe Drinking Water Act, 42 U.S.C. §§ 300f-300j-9 (1982) (the Drinking Water Act), we affirm the rule in all respects and deny the petitions for review.
I. Background
The Drinking Water Act provides the statutory framework for the rule under review. Congress amended the Act in June 1986, after the agency action under review, see Safe Drinking Water Act Amendments of 1986, Pub.L. No. 99-339, 100 Stat. 642 (1986), but, with one exception which we detail below, the amendments do not bear on this case, and we refer in this opinion to the provisions in the pre-1986 version of the Act on which the agency relied. The most important of these provisions for current concerns is section 300g-l. That section requires the Administrator of the EPA to regulate the level of contaminants in drinking water using a three-step process. The first step is the immediate promulgation of “interim primary drinking water regulations.” See 42 U.S.C. § 300g-l(a)(l). In the second step, which is our focus here, the Administrator must establish recommended levels for certain contaminants. To be precise, the Administrator is required to promulgate rules establishing recommended levels “for each contaminant, which in his judgment based on the report [of an independent scientific organization], may have any adverse effect on the health of persons.” Id. § 300g-l(b)(l)(B). If the Administrator determines that a recommended level is necessary for a particular contaminant, “such recommended maximum contaminant level shall be set at a level at which, in the Administrator’s judgment based on such report, no known or anticipated adverse effects on the health of persons occur and which allows an adequate margin of safety.” Id. The recommended levels thus promulgated are nonenforceable health goals. They serve, however, as the benchmark for the third step of the regulatory process — the promul*234gation of maximum contaminant levels (MCLs), which are federally enforceable standards. See NRDC v. EPA, 812 F.2d 721, 723 (D.C.Cir.1987). Under the Drinking Water Act, a MCL must be set “as close to the [recommended level] as is feasible.” 42 U.S.C. § 300g-l(b)(3). Thus, whereas recommended levels are aspirational levels set without regard to practical impediments, MCLs are set at the lowest level feasible, taking into account considerations of cost and available technology and treatment techniques.
EPA began the process that culminated in the rule under review in March of 1982, when the agency published an advance notice of proposed rulemaking for regulation of certain VOCs that had been detected in drinking water. In the notice, the agency requested comment on whether to set recommended levels for carcinogenic VOCs at zero or at “some finite relative risk level.” See 47 Fed.Reg. 9356 (March 4, 1982). In June of 1984, the agency issued a proposed rulemaking, in which it announced a plan to regulate nine of the VOCs listed in the advance notice. The agency proposed recommended levels above zero for the non-carcinogenic VOCs, on the theory that an organism can tolerate and detoxify a certain threshold level of such compounds. For the carcinogenic VOCs, the agency tentatively determined to set the recommended levels at zero, reasoning that any exposure to these compounds would present a risk to human health.
The agency rulemaking here under review followed in November of 1985. The rulemaking assigned recommended levels to eight VOCs according to a three-category scheme. Category I comprised known or probable carcinogens, which the agency concluded should have recommended levels of zero. EPA determined that five of the VOCs properly belonged in this category. One of these five was trichloroethylene, or “TCE.” EPA acknowledged that the evidence of TCE’s carcinogenicity was more equivocal than was the evidence for the other four VOCs in this category, but the agency nevertheless decided to regulate TCE as a probable carcinogen. The EPA placed one VOC, vinylidene chloride, in Category II — VOCs for which there is some equivocal evidence of carcinogenicity. EPA decided not to treat vinylidene chloride as a carcinogen, and it did not set a recommended level of zero for the compound. Rather, the agency decided to establish a recommended level for vinylidene chloride based on the compound’s risk of causing non-cancerous liver and kidney damage. In setting the actual recommended level, however, the agency factored in the equivocal evidence of vinylidene chloride's carcinogenicity. The EPA placed the remaining two VOCs in Category III — contaminants for which there is inadequate or no evidence of carcinogenicity — and assigned recommended levels above zero to these compounds.
These consolidated petitions for review followed. They raise three basic claims. The first is that the agency should not have promulgated recommended levels of zero for all VOCs it considered to be known or probable carcinogens. The second is that the EPA unreasonably characterized TCE as a probable carcinogen. The third claim, brought by NRDC, is that the EPA was required to set a recommended level of zero for vinylidene chloride in light of the agency's finding that the compound is a possible carcinogen. We examine these contentions in turn.
II. Discussion
A. Recommended Levels of Zero for Known or Probable Carcinogens
Faced with the unenviable task of challenging the goal of a total absence of known or probable carcinogens in the nation’s drinking water, industrial petitioners offer two arguments. The first of these is the contention by petitioners Chemical Manufacturers Association and American Petroleum Institute that the EPA’s decision to set recommended levels of zero for the five Category I VOCs was based solely on a misconception that the Drinking Water Act compelled such a result. In support of this claim, petitioners cite language in the order in which the agency refers to a “mandate” from Congress. Petitioners *235also maintain that the agency placed undue reliance on a passage from the Drinking Water Act’s legislative history that states that in cases where there is no safe threshold for a contaminant, the recommended level “should be set at the zero level.” See H.R.Rep. No. 1185, 93rd Cong., 2d Sess. 20 (1974) U.S.Code Cong. & Admin.News 1974, pp. 6454, 6473. In petitioners’ view, the agency considered itself bound to set a recommended level of zero for all known or probable carcinogens, overlooking the possibility that such contaminants do have a tolerably safe threshold within the meaning of the Drinking Water Act.
The record soundly contradicts petitioners’ characterization of the EPA’s decisionmaking. The rule under review noted that the agency had requested comments on three distinct options for setting recommended levels for carcinogens, including nonzero levels based on a calculation of the finite relative risk of each compound. The final rule itself, far from revealing an abdication of judgment, evidences a reasoned determination by EPA that known and probable carcinogens have no safe threshold. The agency wrote, for example
EPA believes that the zero level is necessary to prevent known or anticipated effects from human or probable human carcinogens including a margin of safety. No other margin of safety would be adequate since EPA does not believe a threshold for carcinogens exists.
50 Fed.Reg. at 46,896. The agency did state that “it believed a [recommended level] of zero was more consistent with the [Drinking Water Act] mandate and the legislative history.” Id. at 46,881. The mandate to which this passage refers, however, as the final rule later makes clear, is not a perceived congressional directive to set recommended levels for carcinogens at zero, but rather, “the direction of Congress that EPA set [recommended levels] to prevent known or anticipated effects with a margin of safety.” Id. at 46,896, J.A. 17. The Drinking Water Act clearly does impose this obligation on the agency. Thus, there is no indication that the agency misconstrued or failed to meet its responsibilities under the Drinking Water Act, or that it failed to adequately consider alternatives to the approach it ultimately adopted. Rather, EPA made an expert judgment that there is no safe threshold level for known or probable carcinogens, and set recommended levels of zero for those compounds accordingly.
Petitioners’ second argument is that under the Drinking Water Act, the agency is required to make a predicate finding of “significant risk” before it can regulate a VOC at all. Petitioners contend that the Drinking Water Act must be interpreted to encompass this requirement, because otherwise the statute would confer unfettered discretion to regulate on the EPA, in violation of the delegation doctrine set out by the Supreme Court in A.L.A. Schecter Poultry Corp. v. United States, 295 U.S. 495, 55 S.Ct. 837, 79 L.Ed. 1570 (1935). In advancing this argument, petitioners rely on the Supreme Court’s decision in Industrial Union Department, AFL-CIO v. American Petroleum Institute, 448 U.S. 607, 100 S.Ct. 2844, 65 L.Ed.2d 1010 (1980), popularly known as the “Benzene” decision. In Benzene, the Court construed the Occupational Safety and Health Act (OSHA) as prohibiting the Secretary of Labor from issuing standards to provide safe or healthful employment without a threshold determination “that a place of employment is unsafe — in the sense that significant risks are present and can be eliminated or lessened by changing practices.” Id. at 642, 100 S.Ct. at 2864. Three members of the plurality opinion suggested that one reason for favoring such a construction of OSHA was to avoid any concern of an unconstitutionally broad delegation of legislative power. See Benzene, 448 U.S. at 646, 100 S.Ct. at 2866. Petitioners seize upon this language to claim that the EPA lacked authority to regulate the five Category I VOCs unless it first found that even negligible amounts of the compounds in drinking water presented a significant risk to human health.
The Court based its decision in Benzene on a close reading of the statutory language of OSHA, which we note differs significantly from the statutory scheme *236that we confront in this case. The OSHA language that the Supreme Court interpreted as incorporating a requirement of a finding of significant risk directed the Secretary to set standards “reasonably necessary and appropriate to provide safe or healthful employment.” 29 U.S.C. § 652(8). The Drinking Water Act, by contrast, directs the Administrator to establish a recommended level for “each contaminant which, in his judgment ... may have any adverse effect on the health of persons.” 42 U.S.C. § 300g-l(b)(l)(B) (emphasis added). This language is inconsistent with a requirement that the Administrator make a threshold finding of significant risk; a contaminant may have some adverse effect on the health of persons without posing a significant risk to human health.
Petitioners’ reliance on Benzene is unwarranted not only because of the differences between the two statutory schemes. Whatever the impact of the decision on statutory schemes other than OSHA, Benzene applies only if Congress has not specifically set the agency’s regulatory'agenda. Benzene could not possibly apply in this case because Congress in fact has now told the EPA to regulate the VOCs that are the subject of the rule under review. The 1986 amendments to the Drinking Water Act specifically direct EPA to establish national primary drinking water regulations for 83 enumerated VOCs, including the eight compounds which the EPA regulated in this rule. See Section 101, Safe Drinking Water Act Amendments of 1986, Pub.L. No. 99-339, 100 Stat. 642 (1986). Petitioners point out that the amendments also permit the EPA to substitute a specific contaminant for one of the contaminants enumerated by Congress if, in the judgment of the agency, regulation of the substitute contaminant is more likely to be protective of public health. This added measure of discretion, however, in no way alters the fact that Congress has given a preliminary directive to the EPA to regulate the VOCs at issue here. Thus, we need not determine whether the Benzene decision might have applied by analogy to the unamended Drinking Water Act. Congress now has issued precise marching orders instructing the EPA to regulate these VOCs, and that is all the agency needs to know. See American Mining Congress v. Thomas, 772 F.2d 617, 627-28 (10th Cir.1985) (significant risk determination by EPA inappropriate and unnecessary where Congress had indicated its desire for regulation).
B. Promulgation of a Recommended Level of Zero for TCE
Petitioners Halogenated Solvent Industry Alliance and Diamond Shamrock Chemicals Company protest at length the EPA’s decision to categorize TCE as a probable carcinogen, which resulted in the agency’s promulgation of a recommended level of zero for TCE. Petitioners argue that the EPA miscategorized TCE because any risk presented by the compound is at most de minimis. TCE’s carcinogenicity has been the subject of at least six scientific studies on mice, which have produced varying results. Petitioners belittle the two studies that have indicated that TCE may be a human carcinogen, while they trumpet the studies that point the other way. In brief, petitioners argue that the positive studies are unreliable because they involved suspect dosage levels, dosage methods, TCE grades, mice strains, and mice housing.
Happily, it is not for the judicial branch to undertake comparative evaluations of conflicting scientific evidence. Our review aims only to discern whether the agency’s evaluation was rational. See NRDC v. EPA, 812 F.2d 721, 725 (D.C.Cir.1987). EPA provided ample explanation for its decision to classify TCE as a probable carcinogen. It received and replied to extensive comments on the issue, and it detailed its reasons for giving greater weight to the positive studies than to the negative or inconclusive studies. See 50 Fed.Reg. 46,-886-87. To summarize the EPA’s reasoning, the agency was particularly impressed by finding in two different studies that *237TCE caused a significant increase in the incidence of liver tumors in mice. Although we gather there is some disagreement in the scientific community as to the relevance of evidence of mouse tumors, EPA’s proposed guidelines for carcinogen risk-assessment reasonably take such findings as sufficient evidence of carcinogenicity in the absence of certain contraindications, none of which was present in the studies in question. Moreover, the agency identified concrete flaws in the negative studies. The agency also took note of the consensus among scientists that where cancer is concerned, positive results in one study are not necessarily negated by negative results in another. EPA, in sum, made a reasonable choice to categorize TCE as a probable carcinogen based on a rational evaluation of somewhat conflicting scientific evidence. As we previously have stated, “in an area characterized by scientific and technological uncertainty [w]here administrative judgment plays a key role, ... this court must proceed with particular caution, avoiding all temptation to direct the agency in a choice between rational alternatives.” Environmental Defense Fund v. Costle, 578 F.2d 337, 339 (D.C.Cir.1978). We are satisfied that the alternative for which the agency opted was rational, and we therefore have no cause to disturb the EPA’s choice.
C. Failure to Promulgate a Zero Recommended Level for Vinylidene Chloride
NRDC does not dispute EPA’s finding that vinylidene chloride is a possible — rather than a known or probable — human carcinogen. NRDC instead contends that Congress intended that EPA treat possible carcinogens no differently from known or probable ones. NRDC argues that EPA’s regulatory scheme in effect imposes a requirement that a compound be found to be carcinogenic by a preponderance of the evidence before the agency will establish a recommended level of zero for it. Such a threshold requirement, NRDC contends, violates EPA’s obligation to resolve uncertainty on the side of protecting public health.
We agree with NRDC that a preponderance-of-the-evidence threshold test would probably be inconsistent with Congress' directions in the Drinking Water Act. If the evidence established, for example, a 40% probability that a compound was carcinogenic, the agency’s decision not to regulate would be difficult to square with the Drinking Water Act’s instruction to the agency to establish a recommended level for each contaminant which, in its judgment, may have any adverse effect on health. Such a decision might well constitute an abuse of the discretion the agency is granted under the Drinking Water Act. But that situation in no way describes the instant case, and certainly there is no indication in the final rule that the agency has adopted a general policy not to establish a recommended level for a VOC unless a preponderance of the evidence demonstrates that it is a carcinogen. NRDC perhaps has taken too much to heart the agency’s use of the word “possible” in its categorization of different VOCs. Although that label on its face could augur a preponderance-of-the-evidence test, the agency’s explication of the Category II — compounds for which there is some equivocal evidence of carcinogenicity — makes it clear that the EPA has no such test in mind. Nor does the EPA’s treatment of vinylidene chloride suggest that the agency employed a threshold preponderance-of-the-evidence test. The EPA here reasonably concluded that the evidence of vinylidene chloride’s carcinogenicity was not even close to being in equipoise. The agency pointed out that no fewer than a dozen long-term animal studies had not demonstrated that vinylidene chloride has any carcinogenic effect. See 50 Fed.Reg. 46,888, J.A. 9. Against this data EPA weighed two studies that revealed a possibility of carcinogenic or protocarcinogenic effects, and it noted that the results in both of these studies had limitations that made their applicability to humans highly questionable. The agency therefore had adequate support for its conclusion that the evidence of TCE’s carcinogenicity was sparse and equivocal.
Whether the EPA, having reached such a conclusion, properly de*238dined to set a zero recommended level for vinylidene chloride involves essentially a question of statutory interpretation. The Drinking Water Act provides that the Administrator shall establish recommended levels for each contaminant which, “in his judgment ... may have any adverse effect on the health of persons.” 42 U.S.C. § 300g-l(b)(l)(B). By its terms, the statute grants discretion to the Administrator to determine whether there is sufficient evidence to justify establishing a recommended level for a particular compound. Unless the agency must regulate as a carcinogen every VOC whose carcinogenicity it cannot conclusively disprove, an interpretation that would read the concept of administrative judgment out of the Act, the EPA has discretion not to treat a compound as a carcinogen notwithstanding some equivocal evidence to the contrary. See Environmental Defense Fund v. EPA, 598 F.2d 62, 88 (D.C.Cir.1978) (where evidence of a chemical’s carcinogenicity is inconclusive, Administrator has discretion whether to regulate). EPA’s decision not to establish a recommended level for vinylidene chloride based on the compound’s carcinogenicity thus does not violate the statutory scheme of the Drinking Water Act.
Although EPA did not set a recommended level for vinylidene chloride based on the compound’s carcinogenicity, the agency decided that the compound’s other toxic effects did warrant the establishment of a recommended level. In then calculating the actual recommended level for vinylidene chloride, EPA divided by ten in order to take account of vinylidene chloride’s possibly carcinogenic effects. NRDC argues that it was arbitrary and capricious for the agency to refuse to establish recommended level of zero based on vinylidene chloride’s risk of carcinogenicity but then to take account of the very same risk in setting the recommended level for the compound. We disagree. A careful parsing of the statutory language provides ample support for EPA’s action. The Drinking Water Act provides for promulgation of recommended levels in two steps. In the first, the Administrator determines whether a contaminant may have any adverse effect on the health of persons. If a contaminant may have an adverse effect (for example, in the case of vinylidene chloride, because of its noncarcinogenic risks), the Administrator is directed to set the recommended level at a level at which “no known or anticipated adverse effects on the health of persons occur and which allows an adequate margin of safety.” The statute thus leaves room for the EPA to consider in its actual setting of the recommended level risks other than those that catalyzed the preliminary decision to establish a recommended level. Here, the EPA did just that, concluding that the equivocal evidence of vinylidene chloride’s carcinogenicity, although not sufficient to justify establishing a recommended level on that ground alone, was palpable enough to be accounted for at the “adequate margin of safety” stage. This is neither an unreasonable interpretation of the statute nor an unwise choice of policy.
Finally, NRDC argues that the promulgation of a non-zero recommended level for vinylidene chloride was an unexplained reversal of prior policy, and thus constituted arbitrary and capricious agency action under this court’s decision in Greater Boston Television Corp. v. FCC, 444 F.2d 841 (D.C.Cir.1970), cert. denied, 403 U.S. 923, 91 S.Ct. 2229, 29 L.Ed.2d 701 (1971). NRDC points to EPA’s 1979 promulgation of interim primary drinking water standards for trihalomethanes (THMs). See 44 Fed.Reg. 68,624 (1979). EPA regulated the THMs as carcinogens, having no safe threshold level, on the basis of no more than their structural similarity to a known carcinogen and positive evidence of mutagenicity. NRDC. argues that comparable evidence warrants treatment of vinylidene chloride as a carcinogen in setting recommended levels.
We reject NRDC’s claim because we cannot accept its premise that the EPA had a prior policy under which it would have set a recommended level of zero for vinylidene chloride. The agency’s supposed “policy” *239arose in a proceeding very different from one at issue here, for compounds other than vinylidene chloride. That action scarcely establishes that the agency had a prior policy that would dictate promulgation of a zero recommended level for vinylidene chloride. Indeed, the rule under review was the first time EPA ever promulgated recommended levels for carcinogens under the Drinking Water Act. The agency thus had no prior policy from which to depart, and therefore it was not subject to additional procedural controls set by Greater Boston. In adopting the policy, EPA expressly recognized that it was a new policy in response to developments in cancer risk analysis. See 49 Fed.Reg. 46,-294 (1984).
III. Conclusion
The coincidental opposition to EPA’s rule by public-interest forces and industrial representatives is not relevant to the reviewing process and is not a factor for our consideration, except that it reflects EPA’s careful efforts to carry out its regulatory obligations. The rule under review is measured and well within the regulatory contours established by Congress under the Drinking Water Act. Accordingly, for the foregoing reasons, the petitions for review are denied and the final rule under review is affirmed.
It is so ordered.