Opinion for the court filed by Circuit Judge J. SKELLY WRIGHT.
Concurring opinion filed by Circuit Judge LEVENTHAL.
Circuit Judge MacKINNON filed an opinion concurring in part and dissenting in part on February 28, 1978.
J. SKELLY WRIGHT, Circuit Judge:These cases involve challenges by ASARCO Incorporated, Newmont Mining Corporation, and Magma Copper Company (hereinafter referred to collectively as ASARCO) and the Sierra Club (Sierra)1 to regulations *79issued by the Environmental Protection Agency (EPA).2 The challenged provisions modify previous regulations implementing Section 111 of the Clean Air Act, 42 U.S.C. § 1857c-6 (1970 & Supp. V 1975), as amended, Pub.L.No. 95-95 § 109, 91 Stat. 685, 697-703 (1977), which mandates national emission standards for new stationary sources of air pollution, by introducing a limited form of what the parties call the “bubble concept.” 3 This court has jurisdiction over these petitions under Section 307(b)(1) of the Act, 42 U.S.C. § 1857h-5(b)(1) (1970 & Supp. V 1975), as amended, Pub.L.No. 95-95 § 305, 91 Stat. 772-777.
I
A. Section 111 and the “Bubble Concept”
The 1970 amendments to the Clean Air Act4 were passed in reaction to the failure of the states to cooperate with the federal government in effectuating the stated purposes of the Act, especially the commitment “to protect and enhance the quality of the Nation’s air resources so as to promote the public health and welfare and the productive capacity of its population.” Clean Air Act § 101(b)(1), 42 U.S.C. § 1857(b)(1) (1970). See generally W. Rogers, Environmental Law § 3.1 (1977). The 1970 changes were designed “to improve the quality of the nation’s air,” 84 Stat. 1676 (1970), by increasing the federal government’s role in the battle against air pollution. Train v. Natural Resources Defense Council, Inc., 421 U.S. 60, 64, 95 S.Ct. 1470, 43 L.Ed.2d 731 (1975). The amendments require the states to develop pollution control programs (State Implementation Plans or SIPs) that will keep the levels of given pollutants in the atmosphere below the National Ambient Air Quality Standards (NAAQSs) set by EPA. Clean Air Act §§ 109, 110, 42 U.S.C. §§ 1857c-4, 1857c-5 (1970 & Supp V *801975), as amended, Pub.L.No. 95-95 § 106-108, 91 Stat. 691-697.5
In addition, the 1970 amendments added Section 111, which is the focus of this litigation. This section directs EPA to set specific and rigorous limits on the amounts of pollutants that may be emitted from any “new source” of air pollution. The New Source Performance Standards (NSPSs) established under Section 111 are designed to force new sources to employ the best demonstrated systems of emission reduction.6 Since the NSPSs are likely to be stricter than emission standards under State Implementation Plans, plant operators have an incentive to avoid application of the NSPSs.
The basic controversy in the cases before us concerns the determination of the units to which the NSPSs apply. Under the Act the NSPSs apply to “new sources.” A “new source” is defined as “any stationary source, the construction or modification of which” begins after the NSPS covering that type of source is published. Section 111(a)(2), 42 U.S.C. § 1857c-6(a)(2) (1970) (emphasis added). Further statutory definitions explain the terms used in this one. A “ ‘stationary source’ means any building, structure, facility, or installation which emits or may emit any air pollutant.” Section 111(a)(3), 42 U.S.C. § 1857c-6(a)(3) (1970). A “ ‘modification’ means any physical change in, or change in the method of operation of,, a stationary source which increases the amount of any air pollutant emitted by such source or which results in the emission of any air pollutant not previously emitted.” Section 111(a)(4), 42 U.S.C. § 1857c-6(a)(4) (1970). The statute thus directs that the NSPSs are to apply to any building, structure, facility, or installation which emits or may emit any air pollutant and which is either (1) newly constructed or (2) physically or operationally changed in such a way that its emission of any air pollutant increases.7
The “bubble concept” is based on defining a stationary source as a combination of facilities, such as an entire plant, and applying the NSPSs only when a new plant is constructed or when an existing plant is physically or operationally changed in such a way that net emissions of any pollutant from the entire plant increase. If applied consistently, the bubble concept would allow the operator of an existing plant to avoid application of the strict NSPSs by offsetting any increase in pollution caused by a change in the plant (e.g., modification or replacement of an existing facility, or even addition of a new facility) against a decrease in pollution from other units within the plant as a whole.
B. History of the EPA Regulations
EPA’s original regulations interpreting Section 111, promulgated in 1971, repeated *81the statutory definitions of “stationary source” and “modification” almost word for word8 and did not contain any version of the “bubble concept.” See 36 Fed.Reg. 24877 (1971), codified at 40 C.F.R. §§ 60.2(d) & (h) (1975). EPA acknowledges in its brief that it originally “defined the term new source as ‘an affected facility’ which .in turn was defined as any apparatus to which a standard of performance is specifically applicable.” Brief for EPA at 10 (emphasis added). Affected facilities, and thus new sources, were clearly not synonymous with entire plants. For example, the regulations setting the NSPSs for sources in Portland cement plants identify the following “affected facilities” in such plants: “kiln, clinker cooler, raw mill system, finish mill system, raw mill dryer, raw material storage, clinker storage, finished product storage, conveyor transfer points, bagging and bulk loading and unloading systems.” 40 C.F.R. § 60.60 (1975).9
The “bubble concept” appeared first in proposals from the nonferrous smelting industry and the Department of Commerce (DOC) beginning in December 1972.10 The industry and DOC urged that a stationary source be defined as an entire plant so that no “modifications” of the source would occur unless the total emissions of some pollutant from the plant increased.
In response to industry proposals and demands from DOC, EPA prepared draft regulations in May and July 1974 “clarifying” the previous regulations.11 EPA made some concessions to the industry in these drafts,12 but did not accept the industry’s position that an entire plant should be defined as a single source.13 EPA contin*82ued to resist the bubble concept in meetings with industry representatives, DOC, and the Office of Management and Budget through August 1974. See JA 8-9, 30-34.
Then, in September 1974, the agency again revised its position, making further concessions and proposing new regulations incorporating a limited version of the bubble concept.14 After an additional concession further extending the bubble concept in response to a submission by DOC,15 the proposed regulations were adopted by EPA.16
The new regulations would classify an entire plant as a single stationary source by embellishing the statutory definition of a stationary source as follows:
“Stationary source” means any building, structure, facility, or installation which emits or may emit any air pollutant and which contains any one or combination of the following:
(1) Affected facilities.
(2) Existing facilities.
(3) Facilities of the type for which no standards have been promulgated in this part.
40 C.F.R. § 60.2(d) (1976) (emphasis added). The italicized language is not included in the statutory definition of “stationary source” (“any building, structure, facility, or installation which emits or may emit any air pollutant”), nor was it included in the prior regulations. See 40 C.F.R. § 60.2(d) (1975). Thus the present regulations, instead of limiting the definition of “stationary source” to one “facility” as the statute does, make it cover “any one or combination of” facilities.17 The preamble to the new regulations makes it clear that the purpose of this change is to define a statutory source as an entire plant.18
*83Relying on this new definition of a statutory source, EPA applies the bubble concept to allow a plant operator who alters an existing facility in a way that increases its emissions to avoid application of the NSPSs by decreasing emissions from other facilities within the plant. The regulations provide that “[a] modification shall not be deemed to occur” unless the change in an existing facility results in a net increase in the emission of a pollutant from the whole “source.”19
In spite of strong comments from the nonferrous smelting industry, represented here by ASARCO,20 the new regulations do not consistently apply the bubble concept to treat an entire plant as a single stationary source. EPA continues to apply NSPSs to all newly constructed facilities, even when the emission increases from the new facilities are offset by emission decreases from other facilities in the same plant. Newly constructed facilities are thus treated as independent stationary sources. In order to draw a line defining when the bubble concept will be applied, the regulations classify any changes in existing facilities that cost more than a fixed percentage of the value of the changed facility as “reconstruction.” 40 C.F.R. § 60.15 (1976). “Reconstructed” facilities, like new facilities, are subject to NSPSs regardless ,of whether emissions from the plant of which they are a part increase.
In its petition for review ASARCO argues that the bubble concept must be applied to allow emission increases from reconstruction and new construction to be offset. Sierra argues that the Act defines a “source” as an individual facility, as distinguished from a combination of facilities such as a plant, and that the bubble concept must therefore be rejected in toto. For the reasons stated below we agree with Sierra and remand to EPA for further proceedings consistent with this opinion.
II
A. Scope of Review
The proper scope for judicial review of EPA’s regulations interpreting the Clean Air Act is defined in Section 10(e) of the Administrative Procedure Act, which requires a reviewing court to “decide all relevant questions of law, interpret * * statutory provisions, and * * * hold unlawful and set aside agency action, findings, and conclusions found to be * * * in excess of statutory jurisdiction, authority, or limitations, or short of statutory right * * 5 U.S.C. § 706(2)(C) (1970). See Ethyl Corp. v. EPA, 176 U.S.App.D.C. 373, 541 F.2d 1, 33-34 & n.71 (en banc), cert. denied, 426 U.S. 941, 96 S.Ct. 2663, 49 L.Ed.2d 394 (1976). To be sure, as EPA correctly points out, the Supreme Court and this court have both stated that EPA’s interpretation of the Clean Air Act is to be given considerable deference.21 Nevertheless, it is clear that a reviewing court has the responsibility to examine carefully the words of the statute, the legislative history, and the reasons advanced by the agency to justify its interpretation in order to deter*84mine whether the agency’s interpretation is “sufficiently reasonable that it should [be] accepted by the reviewing courts.” Train v. Natural Resources Defense Council, Inc., supra, 421 U.S. at 75, 95 S.Ct. at 1480.22 Indeed, a panel of this court has recently rejected EPA’s interpretation of another provision of the Clean Air Act. Lubrizol Corp. v. EPA, 183 U.S.App.D.C. 288, 562 F.2d 807 (1977) (finding that even the agency’s persuasive policy arguments were inadequate to extend the commonly accepted meaning of the word “fuel” to include “motor oil” in the absence of any significant support for such an extension in the Act’s legislative history). See also Ass’n of American Railroads v. Costle, 183 U.S.App.D.C. 362, 562 F.2d 1310 (1977) (rejecting the EPA Administrator’s contention that the Noise Control Act of 1972 allowed him discretion not to issue federal noise standards for certain “equipment and facilities” on railroads).
Following the approach defined by these precedents, we first consider the challenge to the regulations raised by the Sierra Club, and then dispose of the challenge raised by ASARCO.
B. Sierra’s Challenge to the Regulations
The Sierra Club’s basic contention is that the new regulations are inconsistent with the plain language of Section 111. The statute defines a stationary source as “any building, structure, facility, or installation which emits or may emit any air pollutant.” Section 111(a)(3), 42 U.S.C. § 1857c-6(a)(3) (1970) (emphasis added). In contrast, the new regulations define stationary source to include “any * * * combination of* * facilities * * 40 C.F.R. § 60.2(d) (1976) (emphasis added).
This change in the definition of a stationary source is essential to EPA’s adoption of the bubble concept. By treating a combination of facilities as a single source, the regulations allow a facility whose emissions are increased by alterations to avoid complying with the applicable NSPS as long as emission decreases from other facilities within the same “source” cancel out the increase from the altered facility.23 Sierra argues forcefully that this result is incompatible with the statute’s mandate that NSPSs should be applied to “any structure, building, facility, or installation” that undergoes “any physical change * * * or * * * change in the method of operation * * * which increases the amount of any air pollutant emitted by such [structure, building, facility, or installation].” 42 U.S.C. §§ 1857c-6(a)(3), 1857c-6(a)(4) (1970) (emphasis added). See brief for petitioner Sierra Club at 25-33.
EPA responds that the “broad” statutory definition of stationary source gives it “discretion” to define a stationary source as either a single facility or a combination of facilities. Brief for EPA at 13-16. We find this response unpersuasive.24 The reg*85ulations plainly indicate that EPA has attempted to change the basic unit to which the NSPSs apply from a single building, structure, facility, or installation — the unit prescribed in the statute — to a combination of such units. The agency has no authority to rewrite the statute in this fashion. See, e. g., Ass’n of American Railroads v. Costle, supra; Lubrizol Corp. v. EPA, supra.
Our conclusion that the regulations incorporating the bubble concept must be rejected as inconsistent with the language of the Act is reinforced when we consider the purpose of the Clean Air Act and Section 111, the confusion generated by the present regulations, and the weakness of EPA’s arguments in favor of the bubble concept.
“[T]he goal of the Clean Air Act,” as EPA admits in its brief, “is to enhance air quality and not merely to maintain it.” Brief for EPA at 17 (emphasis added). See Clean Air Act § 101(b)(1), 42 U.S.C. § 1857(b)(1) (1970). Section Ill’s provisions mandating New Source Performance Standards were passed because Congress feared that the system of state plans designed to keep air pollution below nationally determined levels was insufficient by itself to achieve the goal of protecting and improving air quality.25 The New Source Performance Standards are designed to enhance air quality by forcing all newly constructed or modified buildings, structures, facilities, or installations to employ pollution control systems that will limit emissions to the level “achievable through application of the best technological system of continuous emission reduction which * * the Administrator determines has been adequately demonstrated.” 42 U.S.C. § 1857c-6(a)(1) (1970), as amended, Pub.L.No. 95-95 § 109(c)(1), 91 Stat. 699-70026 The bubble *86concept in the challenged regulations would undercut Section 111 by allowing operators to avoid installing the best pollution control technology on an altered facility as long as the emissions from the entire plant-do not increase. For example, under the bubble concept an operator who alters one of its facilities so that its emission of some pollutant increases might avoid application of the NSPS by simultaneously equipping other plant facilities with additional, but inferior, pollution control technology or merely reducing their production.27 Applying the bubble concept thus postpones the time when the best technology must be employed and at best maintains the present level of emissions.
Moreover, the challenged regulations are internally inconsistent and create confusion by defining a stationary source one way (as an entire plant) when determining whether a “source” has been “modified,” and another way (as an individual facility) when determining whether a “source” has been newly constructed or “reconstructed.”28 This inconsistency is apparently the result of a “compromise” between EPA’s original regulations, which followed the Act in treating a single facility as a source, and the industry position presented by ASARCO that a source must be defined as an entire plant.29 We are unable to understand why EPA should find it necessary to compromise by adopting a position that it admits is contrary to both the language and the basic purpose of the Act.30
EPA’s main argument in support of its regulations is that its version of the bubble concept is necessary to provide flexibility in applying the NSPSs to modified facilities because the cost of bringing existing facilities into compliance with NSPSs is allegedly much greater than the cost of bringing new facilities into compliance.31 This argument does not survive analysis. The record does not show that any version of the bubble concept is needed to provide flexibility to the operators of existing facilities. Under provisions of the regulations that are not challenged in this litigation, the operator of an existing facility can make any alterations he wishes in the facility without becoming subject to the NSPS as long as the level of emissions from the altered facility does not increase.32 Thus the level of emissions before alterations take place, rather than the strict NSPS, effectively defines the standard that an altered facility *87must meet. The record does not indicate why more flexibility than this is necessary or even appropriate. Even if flexibility were a problem, the statute on its face allows for cost considerations to be taken into account in setting NSPSs, rather than in determining whether the standards will apply to whole plants or to individual facilities within those plants.33
Finally, the record indicates that the bubble concept has been supported by examples drawn from circumstances peculiar to the nonferrous smelting industry.34 As EPA itself recognizes in its brief,35 the proper place to consider the problems a particular industry will have in meeting the NSPSs is in proceedings dealing with the standards for that particular industry, not in regulations setting standards for all industries.36
We therefore agree with the Sierra Club that EPA’s regulations incorporating the bubble concept are inconsistent with the language and purpose of the statute and cannot be justified by any alleged need for flexibility.
C. ASARCO’s Challenge to the Regulations
The dispute between ASARCO and EPA centers on how far the bubble concept should extend. ASARCO asserts that a stationary source must be defined as an entire plant for all purposes and that the NSPSs should therefore never apply to an existing plant, even if new facilities are built or old ones are “reconstructed,” unless the net emissions of some pollutant from the entire plant increase.37 ASARCO argues that the inconsistency in EPA’s present definition of a stationary source38 — an inconsistency created by EPA’s partial concessions to ASARCO’s position — should be resolved39 by doing away with all limitations on the bubble concept. Brief for petitioner ASARCO at 18-23.
Since we find that any version of the bubble concept is incompatible with the language of the Act and contrary to its purpose, ASARCO’s position is clearly untenable.40 Our holding therefore resolves the *88inconsistency in the challenged regulations by eliminating its source: the bubble concept.
Accordingly, we remand to the EPA for further proceedings not inconsistent with this opinion.
So ordered.
. As a preliminary'matter the petitioners in No. 76-1030, as intervenors in No. 76-1037, have made á motion, not joined by the EPA, that Sierra’s petition be dismissed because Sierra did not participate in the informal rulemaking proceedings and because the points raised in Sierra’s petition were allegedly not adequately presented by the parties that did participate.
In fact, the issue raised by Sierra’s petition— whether the “bubble concept” adopted by the new regulations (see 188 U.S.App.D.C. at 80, 578 F.2d at 322 infra) is consistent with the language of the Clean Air Act — was a central issue in EPA’s rulemaking proceedings to determine whether its existing regulations should be changed to accommodate the bubble concept. That issue was raised by those participating in. the rulemaking proceedings, just as it is raised by Sierra and ASARCO in this litigation. See, e.g., Meeting Report: Meeting on Definition of Modification, EPA/Nonferrous Smelter Industry, Sims L. Roy, Jr. to Files, July 5, 1973, at 1, JA 8 (hereinafter cited as Meeting Report, July 5, 1973) (“EPA indicated that the legality of the proposed industry definition of modification [adopting the bubble concept] would be considered”) (emphasis added); Public Comment Summary: Modification, Notification, Reconstruction, Revised April 1976, at 12-13, JA 110-111 (response to Comment 20). The final report to the EPA Administrator on the regulations noted that “[i]t appears that the most probable area of controversy which would result from the promulgation of this package will be the ‘bubble concept’ * * Rulemaking to Clarify the Modification Provisions of Section 111 of the Clean Air Act — ACTION MEMORANDUM, Nov. 24, 1975, at 4, JA 97. The issue raised by Sierra was thus not only raised and considered in the proceedings below; it was a substantial part of the fundamental issue in those proceedings. See 40 Fed. Reg. 58416-58417 (1975), JA 99-100 (EPA’s *79preamble to regulation discussing “Terminology” and “The ‘Bubble Concept’ ”). We therefore find no merit in intervenors’ motion to dismiss.
Even if this point were not dispositive, other factors in this case would lead us to exercise our discretion to deny the motion to dismiss. See, e.g., NLRB v. Industrial Union of Marine & Shipbuilding Wkrs., 391 U.S. 418, 426 n.8, 88 S.Ct. 1717, 20 L.Ed.2d 706 (1968); Hayes v. Secretary of Defense, 515 F.2d 668, 673-675 (1975); Natural Resources Defense Council, Inc. v. Train, 510 F.2d 692, 703 (1974); K. Davis, Administrative Law of the Seventies § 20.01 (1976). First, EPA itself did not move for dismissal on grounds of exhaustion or join in intervenors’ motion. The agency’s failure to insist upon exhaustion in this particular case suggests that, in EPA’s view, the costs — in terms of interfering with agency proceedings, creating an incentive to ignore agency processes in the future, and depriving the agency of a greater opportunity to apply its expertise or of a chance to “correct its own errors” —do not outweigh the benefits of this court’s immediate consideration of Sierra’s petition. See K. Davis, Administrative Law of the Seventies § 20.00 (1977 Supp.); cf. Weinberger v. Salfi, 422 U.S. 749, 764-767, 95 S.Ct. 2457, 45 L.Ed.2d 522 (1975). Second, the issues raised by Sierra’s petition are essentially issues of statutory interpretation, which depend primarily on the words of the statute and the relevant legislative history, rather than on a record of fact-finding developed by an agency with special expertise. See McKart v. United States, 395 U.S. 185, 197-199, 89 S.Ct. 1657, 23 L.Ed.2d 194 (1969). Third, since the petitions of ASARCO and Sierra raise essentially the same question — whether the “bubble concept” regulations are consistent with the mandate of the Clean Air Act — the interests of judicial efficiency are served by hearing and deciding these cases together.
As a final consideration, we note that Congress has now amended § 307 of the Clean Air Act to require a form of exhaustion of administrative remedies as a prerequisite to judicial review under that section in the future. See Pub.L.No. 95-95 § 305(a), 91 Stat. 775 (1977) (adding § 307(d)(7)(B)). The amendments specifically provide, however, that this exhaustion requirement will apply only to rules proposed “after ninety days after the date of enactment” of the amendments, which were enacted on August 7, 1977. See Pub.L.No. 95-95 § 305(a), 91 Stat. 775-776 (adding § 307(d)(ll)). This provision indicates that Congress wished exhaustion questions related to rules proposed before that date to be determined according to existing law.
. 40 Fed.Reg. 58416 (1975), JA 99, codified at 40 C.F.R. §§ 60.2(d), (h), (aa), (bb); 60.5; 60.7; 60.14; 60.15 (1976). The particular portions of these regulations being challenged are indicated in the discussion below.
. For an explanation of the “bubble concept” see 188 U.S.App.D.C. at 80, 578 F.2d at 322 infra.
. Pub.L.No. 91-604, 84 Stat. 1676 (1970).
. If a state does not submit an adequate SIP within a specified time limit, EPA is authorized to design and implement its own plan. 42 U.S.C. § I857c-5(c) (Supp. V 1975), as amended, Pub.L.No. 95-95 § 108(d), 91 Stat. 685, 694-695. Federal and state authorities are given joint powers to enforce the plans. See, e.g., 42 U.S.C. § 1857C-9 (1970 & Supp. V 1975), as amended, Pub.L.No. 95-95 § 113, 91 Stat. 709.
. The 1970 amendments provided that the NSPSs were to “reflect!] the degree of emission limitation achievable through the application of the best system of emission reduction which (taking into account the cost of achieving such reduction) the Administrator determines has been adequately demonstrated.” Section 111(a)(1), 42 U.S.C. § 1857c-6(a)(l) (1970).
This language was amended in 1977, but the new provisions still reflect a commitment to requiring the best technology. The NSPSs must now “reflect the degree of emission limitation and the percentage reduction achievable through the application of the best technological system of continuous emission reduction which (taking into consideration the cost of achieving such emission reduction, any nonair quality health and environmental impact and energy requirements) the Administrator determines has been adequately demonstrated.” Pub.L.No. 95-95 § 109(c)(1)(A), 91 Stat. 699-700.
.EPA is instructed to publish a list of categories of stationary sources and to issue regulations setting NSPSs for the sources within these categories. 42 U.S.C. § 1857c-6(b)(l) (1970 & Supp. V 1975), as amended, Pub.L.No. 95-95 §§ 109(c)(2), 109(c)(3), 91 Stat. 700-701. The 1977 amendments add provisions designed to expedite promulgation of NSPSs for all “major stationary sources.” Pub.L.No. 95-95 § 109(a), 91 Stat. 697-699.
. The definition of “modification” in the regulations differed in several respects from the statutory definition. The regulations specified that “[rjoutine maintenance, repair, and replacement” would not constitute “physical changes” and that “change[s] in the method of operation” would not include an increase in the production rate up to the “operating design capacity” of a facility, “[an] increase in the hours of operation,” or “[u]se of an alternative fuel or raw material” that the facility was previously designed to accommodate. 40 C.F.R. § 60.2(h) (1975). These provisions have been slightly amplified and carried over into the new regulations, 40 C.F.R. § 60.14(e) (1976), but they are not challenged by the petitioners here. See br. for petitioner Sierra Club at 7 & n.4.
. An entire plant is an awkward unit to set standards for. Indeed, even under the new regulations EPA sets standards for facilities rather than entire plants. See, e.g., 40 C.F.R. §§ 60.2(d), 60.120-60.122 (1976). A single standard applies to an entire plant only when the plant contains a single facility. See, e.g., 40 C.F.R. § 60.90 (1976) (NSPS for asphalt concrete plants). Even ASARCO does not argue that EPA should set a single NSPS for an entire plant. See br. for petitioner ASARCO at 6-7; reply br. for ASARCO at 3-4. Yet the Act’s direction that NSPSs are to be set “for sources" indicates that the units for which standards are set are sources. 42 U.S.C. § 1857c-6(b)(l)(B) (Supp. V 1975), as amended, Pub.L.No. 95-95 § 109(c)(2), 91 Stat. 700 (emphasis added).
. See, e.g., letter from James M. Henderson of ASARCO to Donald F. Walters, then Chairman of the National Air Pollution Control Techniques Advisory Committee, Dec. 27, 1972, at 4 (incorporated by reference in letter from David W. Miller to Don R. Goodwin (EPA), Nov. 27, 1974, Doc. No. 35, at 3); telegram from Dept, of Commerce to Dr. Bernard Steigerw'áld (EPA), Mar. 3, 1973, Doc. No. 133, at 1; Meeting Report, July 5, 1973, supra note 1, JA 8; letter from George Wunder (Anaconda Co.) to Don R. Goodwin (EPA), Feb. 7, 1974, JA 12.
. See Minutes of Meeting: National Air Pollution Control Techniques Advisory Committee, May 21-22, 1974, Doc. No. 125, Appendix F (hereinafter cited as May Meeting Minutes); br. for petitioner Sierra Club at 10 & n.7.
. The draft regulations defined a single “stationary source” as containing a “combination of facilities.” See May Meeting Minutes, supra note 11, Appendix F at 13. At this stage, however, a “combination of facilities” was not intended to refer to an entire plant. EPA intended to “limit[] the scope of a combined emission system to a group of facilities whose emissions can be controlled by an existing common control system.” Id. at 13. Furthermore, the combined emissions approach was to be applied only to “those facilities where a high-quality control system already exists and where substantial .savings of control costs or energy could be achieved.” Id.
. EPA’s major objection to the bubble concept was that it would make emission standards extremely difficult to enforce. See, e.g., May Meeting Minutes, supra note 11, at 13; Draft Memorandum, “Proposed Rulemaking to Clarify the Modification Provisions of Section 111 of the Clean Air Act — ACTION MEMO,” July 12, *821974, Tab D at 2. The agency was also concerned that an approach based on maintaining existing levels of pollution would reward those operators who were presently using the fewest controls and use up air that would otherwise be available for new construction complying with NSPS. See, e.g., Minutes of Meeting: National Air Pollution Control Techniques Advisory Committee, Jan. 8-10, 1974, Doc. No. 130, at 7-8; Public Comment Summary, Revised April 1976, at 12-14, JA 110-112 (response to Comment 24).
. See Proposed Rulemaking to Clarify the Modification Provisions of Section 111 of the Clean Air Act — ACTION MEMO., with Tabs A, B, C, and D, Sept. 30, 1974, JA 45-62; 39 Fed.Reg. 36946-36950 (1974), JA 63-67.
. See letter from William C. Rountree, Assistant General Counsel, DOC, to Alvin L. Aim, Assistant Administrator for Planning and Management, EPA, Dec. 9, 1974, JA 91-93. The proposed regulations permitted emission increases from altered facilities to be offset only by emission decreases from other facilities of the type for which NSPSs were prescribed. This limitation was considered necessary for accurate measurement. It was dropped in the final regulations. See Rulemaking to Clarify the Modification Provisions of Section 111 of the Clean Air Act — ACTION MEMORANDUM, Nov. 24, 1975, at 2, JA 95; 40 C.F.R. § 60.14(d) (1976).
. 40 Fed.Reg. 58416 (1975), JA 99.
. The EPA’s definition of a “facility,” which this court accepts, is “any apparatus to which a standard of performance is specifically applicable.” See 188 U.S.App.D.C. at 81, 578 F.2d at 323 supra. This definition is clearly designed to designate as “facilities” those units of equipment — be they individual machines, combinations of machines, or even entire plants— that the agency finds to be appropriate units for separate emission standards. A cursory review of the EPA’s regulations indicates that the units designated as “facilities” under this definition are usually larger than individual machines or single pieces of equipment, and are sometimes whole plants. See text and note at note 9 supra. In designating what will constitute a facility in each particular industrial context, EPA is guided by a reasoned application of the terms of the statute it is charged to enforce, not by an abstract “dictionary” definition. This court would not remove this appropriate exercise of the agency’s discretion. See Union Electric Co. v. EPA, 427 U.S. 246, 256, 96 S.Ct. 2518, 49 L.Ed.2d 474 (1976); Train v. Natural Resources Defense Council, Inc., 421 U.S. 60, 75, 87, 95 S.Ct. 1470, 43 L.Ed.2d 741 (1975); Ethyl Corp. v. EPA, 176 U.S.App.D.C. 373, 541 F.2d 1, 12 n.16, 31 n.64 (en banc), cert. denied, 426 U.S. 941, 96 S.Ct. 2663, 49 L.Ed.2d 394 (1976).
.Generally speaking, “sources” are entire plants, while “facilities” are identifiable pieces of process equipment or individual components which when taken together would comprise a source. * * *
40 Fed.Reg. 58416 (1975), JA 99.
.A modification shall not be deemed to occur if an existing facility undergoes a physical or operational change where the owner or operator demonstrates to the Administrator’s satisfaction * * * that the total emission rate of any pollutant has not increased from all facilities within the stationary source to which appropriate reference, equivalent, or alternative methods * * * can be applied. An owner or operator may completely and permanently close any facility within a stationary source to prevent an increase in the total emission rate regardless of whether such reference, equivalent or alternative method can be applied, if the decrease in emission rate from such closure can be adequately determined by any of the procedures prescribed under paragraph (b) of this section. The owner or operator of the source shall have the burden of demonstrating compliance with this section.
40 C.F.R. § 60.14(d) (1976).
. See, e.g., Comments of American Smelting & Refining Co., Newmont Mining Corp., and Magma Copper Co. on Proposed Rules, Nov. 27, 1974, at 42-50, JA 79-87.
. See Union Electric Co. v. EPA, supra note 17, 427 U.S. at 256, 96 S.Ct. 2518; Train v. Natural Resources Defense Council, Inc., supra note 17, 421 U.S. at 75, 87, 95 S.Ct. 1470; Ethyl Corp. v. EPA, supra note 17, 176 U.S.App.D.C. 373, 541 F.2d at 12 n.16, 31 n.64.
. See, e.g., id., 421 U.S. at 75-98, 95 S.Ct. 1470; Union Electric Co. v. EPA, supra note 17, 427 U.S. at 256-266, 96 S.Ct. 2518; Ethyl Corp. v. EPA, supra note 17, 541 F.2d at 11-33.
. See text and note at note 19 supra.
. The Act’s definition of a “stationary source” as “any building, structure, facility, or installation which emits or may emit any air pollutant” is a broad and comprehensive one that cannot be reconciled with the regulations’ restrictive definition of a source as an entire plant. In its briefs as intervenor and petitioner ASARCO attempts to justify the definition of a stationary source as an entire plant by pointing to passages in the legislative history that contain lists of the types of polluters Congress sought to regulate: e. g., “refineries, steel mills, primary smelting plants * * S.Rep.No. 91-1196, 91st Cong., 2d Sess. 16, in Senate Comm, on Pub. Works, 91st Cong., 2d Sess., 1 Legislative History of the Clean Air Act Amendments of 1970, at 416 (Comm.Print 1974) (hereinafter cited as Legislative History); “power plants, steel mills, and cement plants * * 116 Cong.Rec. 42384 (1970) (remarks of Sen. Muskie), in 1 Legislative History at 130. Br. for petitioner ASARCO at 22. EPA itself, however, points out that the legislative history contains lists mentioning “numerous other terms which refer to operations rather than entire plants as physical entities.” Br. for EPA at 14 & n.12. Sierra argues further that these lists were not intended to be exclusive, that these larger “sources” are in fact regulated when their components are regulated, and that the Senators mentioned whole plants as sources because they lacked the familiarity with various industries to mention smaller units. Br. for amicus curiae Sierra Club at 16-17. We find *85the legislative history on this point a much less reliable guide than the words of the statute itself.
EPA attempts to base its “discretion” to redefine the term “stationary source” on § 111(b)(2), 42 U.S.C. § 1857c-6(b)(2) (1970):
The Administrator may distinguish among classes, types, and sizes within categories of new sources for the purpose of establishing [New Source Performance] standards.
This language on its face merely allows the Administrator to set different standards for different classes, types, and sizes of sources. It does not give the Administrator authority to rewrite the definition of a stationary source, or to use different and inconsistent definitions when considering modification and new construction.
Finally, EPA claims that § 111 is aimed primarily at new construction, br. for EPA at 19-20, and candidly admits that it intended “to develop a modification regulation * * * which minimizes that impact on existing sources.” May Meeting Minutes, supra note 11, at 11. The Act’s language, however, is aimed at new sources, not just new construction, and defines existing sources that are altered so that their emissions increase as new sources.
. For example, the National Ambient Air Quality Standards will not encourage a state to adopt measures to improve air quality unless existing air pollution in the state is above the national limits. Congress feared that states with presently low levels of pollution would have an incentive to adopt lenient State Implementation Plans to attract industry until pollution reached the national limits. “Forum shopping” by industry on the basis of the leniency of SIPs would then cause all air to deteriorate to the minimum quality set by the national standards. See, e. g., 116 Cong.Rec. 32902 (1970) (remarks of Sen. Muskie) in 1 Legislative History, supra note 24, at 227. Congress was also concerned that the pollution limits would have an adverse effect on industrial growth since no new plants could be built if existing industry was already polluting up to the national limits. See, e. g., S.Rep.No. 1196, 91st Cong., 2d Sess. 17 (1970), in 1 Legislative History, supra note 24, at 417.
. See, e. g., S.Rep.No. 91-1196, 91st Cong., 2d Sess. 15-17 (1970), in 1 Legislative History, supra note 24, at 415-417:
The provisions for new source performance standards are designed to insure that new stationary sources are designed, built, equipped, operated, and maintained so as to reduce emissions to a minimum. * * * The maximum use of available means of preventing and controlling air pollution is essential to the elimination of new pollution problems while cleaning up existing sources.
* * ifi * *
Standards of performance should provide an incentive for industries to work toward constant improvement in techniques for preventing and controlling emissions from stationary sources, since more effective emission control will provide greater latitude in the selection of sites for new facilities. * *
. See 40 C.F.R. § 60.14(d) (1976).
. See 188 U.S.App.D.C. at 82, 578 F.2d at 324-325 supra.
. See, e. g., Rulemaking to Clarify the Modification Provisions of Section 111 of the Clean Air Act — ACTION MEMORANDUM, Nov. 24, 1975, at 2, JA 95 (regulations described as a “compromise” between consistently treating entire plants as single sources and rejecting the bubble concept).
. See id. (the only reason given for not rejecting the bubble concept entirely is that this “would meet strong opposition from the smelter industry and the Department of Commerce”). In its brief EPA makes convincing arguments against adopting ASARCO’s version of the “bubble concept.” See br. for EPA at 24-38. The agency fails to explain adequately, however, why the same arguments do not undercut its own version of the bubble concept.
. See br. for EPA at 18-21. EPA also claims that its regulation on “reconstruction,” 40 C.F.R. § 60.15 (1976), supports the reasonableness of the modified bubble concept by protecting against possible abuses. Br. for EPA at 21-22. We note that the absence of any provision in these regulations for aggregating expenditures on existing equipment over the years creates a troublesome loophole in this “protection.” See reply br. for petitioner Sierra Club at 13-15. Furthermore, though these regulations make EPA’s version of the bubble concept less damaging to the purposes of the Act, they do not explain why the concept in any form is a reasonable approach under the statute.
.See 40 C.F.R. §§ 60.2(h), 60.14(a) (1976); 39 Fed.Reg. 36946 (1974); Proposed Rulemaking to Clarify the Modification Provisions of Section 111 of the Clean Air Act — ACTION MEMO., Sept. 30, 1974, Tab A, at 3-4, JA 51-52 (explains EPA decision to determine whether an increase in emissions has taken place by measuring actual rather than potential —precontrolled—emissions); Public Comment Summary; Modification, Notification, Reconstruction, Revised April 1976, at 11, JA 109 (response to Comment 20). Since these regulations are not challenged here, we express no opinion on their validity.
. See 42 U.S.C. § 1857c-6(a)(l) (1970), as amended, Pub.L.No. 95-95 § 109(c)(1)(A), 91 Stat. 699-700, quoted at note 6 supra.
. Rulemaking to Clarify the Modification Provisions of Section 111 of the Clean Air Act— ACTION MEMORANDUM, Nov. 24, 1975, at 1, JA 94.
. Br. for EPA at 33-34.
. The effect of our decision is thus to require the parties to re-address their concerns in proceedings more appropriate for considering concrete applications of the Act’s provisions as properly interpreted. The industry petitioners in this action have already filed a petition to review the regulations setting NSPSs for the nonferrous smelting industry. See ASARCO, Inc., Newmont Mining Corp. and Magma Copper Co. v. EPA, D.C.Cir. No. 76-1093, filed Jan. 29, 1976.
. Br. for petitioner ASARCO at 23-31.
. See 188 U.S.App.D.C. at 85, 578 F.2d at '327-328.
. ASARCO’s position itself is not consistent. If an entire plant were consistently defined as a single stationary source, the whole plant would become subject to NSPSs whenever any alterations in the plant, e. g., addition of a new facility, caused a net increase in the emission of any pollutant from the plant. Yet ASARCO supports EPA’s regulation that prevents this result. See 40 C.F.R. § 60.14(c) (1976); reply br. for petitioner ASARCO at 3-4.
.ASARCO claims that adopting its proposal could lead in some instances to preventing short-term increases in pollution that would not be prevented if a “source” is defined as a single facility. See br. for petitioner ASARCO at 14-17. Essentially, the argument is that under the bubble concept an operator will have an incentive to keep plant emissions constant, while without the bubble concept emissions would increase at least by some amount whenever a new facility is added to the plant. EPA responds that the argument is based on special conditions present in the smelting industry and that in any event adopting ASARCO’s proposal would “significantly increase emissions over the long-term and reward existing sites that have the highest pollution levels.” Br. for EPA at 34.
Indeed, the consequences that would follow if ASARCO’s position were adopted indicate the dangers inherent in accepting any version of the bubble concept. Treating whole plants as single sources would grant the operators of existing plants permanent easements against federal new source standards — and the worst polluters would get the largest easements. The “best” pollution control technology would nev*88er have to be installed, and existing plants would use up much of the available clean air, inhibiting construction of new plants (which would comply with the NSPSs). As EPA explained:
If the bubble concept were extended to cover new construction, large sources of air pollution could avoid the application of new source performance standards indefinitely. Such sources could continually replace obsolete or worn out facilities with new facilities of the same type. If the same emission controls were adopted, no overall emission increase would result. In this manner, the source could continue indefinitely without ever being required to upgrade air pollution control systems to meet standards of performance for new facilities. * * *
40 Fed.Reg. 58416, 58417 (1975), JA 99, 100. Unfortunately, these fears did not deter EPA from adopting regulations that would cause the same problem on a lesser scale.