Opinion for the Court filed by Chief Judge WALD.
Dissenting opinion filed by Circuit Judge BORK.
WALD, Chief Judge:
This case concerns the availability and scope of judicial review of the National Highway Transportation Safety Administration’s (NHTSA) decision to deny the Center for Auto Safety’s (CAS)1 petition to reopen an enforcement investigation against the Ford Motor Company for safety defects in automobiles built between 1966 and 1979. The National Traffic and Motor Vehicle Safety Act of 1966,2 as amended, states that if the Secretary of Transportation makes a final determination that a safety-related defect exists, the Secretary of Transportation “shall order the manufacturer [to recall and remedy] such defect....” 15 U.S.C. § 1412(b).3 The Act also provides that “[a]ny interested person” may petition the Secretary “to commence a proceeding to determine whether to issue a[] [final] order pursuant to section 1412(b).” Id. § 1410a(a). The NHTSA regulations provide that the agency will grant a petition if the agency finds that there is a “reasonable possibility” of a safety-related defect in the manufacturers’ cars. 49 C.F.R. § 552.8 (1987). NHTSA’s [801]*801regulations also state the decision to grant or deny a petition is to be based on a “technical review” of evidence relevant to the safety issue. Id, § 552.6. In this case, the CAS challenges the Secretary’s denial of its petition to reopen as arbitrary and capricious. The District Court found that the Secretary’s decision was unreviewable under Heckler v. Chaney, 470 U.S. 821, 105 S.Ct. 1649, 84 L.Ed.2d 714 (1985), but, even if reviewable, that the scope of review under Dunlop v. Bachowski, 421 U.S. 560, 95 S.Ct. 1851, 44 L.Ed.2d 377 (1975), must be limited to an examination of the agency’s stated reasons for its decision.
We hold that denials of petitions to investigate alleged safety defects under 15 U.S.C. § 1410a and 49 C.F.R. § 552.8 are subject to judicial review under the Administrative Procedure Act (APA) to assure (1) that NHTSA acts according to its own “reasonable possibility” standard and (2) that its finding of no “reasonable possibility” is not “arbitrary” or “capricious.” 5 U.S.C. § 706(2)(A). We also hold that in order to conduct review of the agency’s finding on the likelihood of a safety-related defect, the court must examine not merely the statement of reasons given by the Administrator, but the evidence compiled in the “technical review,” on which the agency relies in making its decision. Consequently, we reverse the decision of the District Court denying and limiting review, and remand for a decision based on the usual standards of review under the APA.
I. Background
A. Statutory and Regulatory Framework
The Motor Vehicle Safety Act establishes certain standards governing the treatment of citizen petitions to investigate alleged safety defects:
Within 120 days after the filing of a petition ..., the Secretary shall either grant or deny the petition. If the Secretary grants such petition, he shall promptly commence the proceeding requested in the petition. If the Secretary denies such petition he shall publish in the Federal Register his reason for such denial.
Id. § 1410a(d). Although the Act does not lay down any specific procedure or standard for making petition decisions, it does state that “[t]he Secretary may hold a public hearing or may conduct such investigation or proceeding as he deems appropriate in order to determine whether or not such petition should be granted.” Id. § 1410a(c).
NHTSA has, however, itself promulgated a detailed set of regulations which “establishes procedures for the submission and disposition of petitions filed by interested persons pursuant to [15 U.S.C. § 1410a].” 49 C.F.R. § 552.1 (1987). In relevant part, these regulations stipulate that “[t]he appropriate Associate Administrator conducts a technical review of the petition, to determine whether there is a reasonable possibility that the requested order will be issued at the conclusion of the appropriate proceeding.” Id. § 552.6.
This “technical review” is based on evidence available to NHTSA from one or more potential sources: the petition itself, “information already in the possession of the agency,” “the collection of additional information,” or the collection of evidence at a “public meeting.” Id. Then:
At the conclusion of the technical review, the Administrator or his delegate determines whether there is a reasonable possibility that the order requested in the petition will be issued at the conclusion of the appropriate proceeding. If such reasonable possibility is found, the petition is granted. If is it not found the petition is denied.
Id. § 552.8 (emphasis added). Under this regulation, it is undisputed that the petition to open an investigation must be granted if “there is a reasonable possibility that [a safety-related defect exists],”4 and that the [802]*802“reasonable possibility” determination is to be based on an analysis of the evidence compiled in the “technical review.”
B. The Relevant Facts of This Case
This appeal concerns alleged defects in automatic transmissions built between 1966 and 1979 by the Ford Motor Company. Appellants claim that the defects cause Ford cars to disengage from “Park” and roll without warning. On March 6, 1985, they petitioned NHTSA to open an investigation that would lead to an order requiring Ford to remedy the defective transmissions. An earlier NHTSA investigation of similar allegations had been terminated without a final determination on the existence of a defect, when NHTSA and Ford entered a settlement agreement that required Ford to notify owners about the possibility of a defect. Under the settlement agreement, however, NHTSA explicitly reserved ■ the right to commence a new proceeding on the alleged defect if additional facts warranted such action. See Center for Auto Safety v. Lewis, 685 F.2d 656, 661 n. 5 (D.C.Cir.1982).5 In their March 1985 petition, appellants claimed to have such additional evidence as to warrant the opening of a new investigation on the matter.
On July 12, 1985, the NHTSA Administrator denied appellants’ petition, stating that her review of the evidence in the administrative record, including appellant’s petition, new information received from Ford and other car manufacturers, and internal NHTSA accident reports, had failed to “convinced” her “that a final defect determination is now warranted or even likely if further investigation were undertaken.” Appellee’s Appendix at 35. It is that determination which CAS unsuccessfully sought to have reviewed in the District Court.
II. Availability of Judicial Review
Under the Administrative Procedure Act, NHTSA’s denial of appellant’s petition is subject to judicial review unless Congress has “affirmatively precluded review” in the Motor Safety Act or unless NHTSA’s decision is not governed by a “meaningful” legal standard against which a court can measure its validity. Heckler v. Chaney, 470 U.S. 821, 830, 105 S.Ct. 1649, 1655, 84 L.Ed.2d 714 (1985). The first exception to APA review, affirmative preclusion by Congress, derives from 5 U.S.C. § 701(a)(1); the second, lack of judicially manageable standards, derives from § 701(a)(2).6 In Chaney, the Supreme Court observed that “an agency decision not to enforce” usually is not governed by “judicially manageable standards” and “often involves a complicated balancing of a number of factors which are peculiarly within [the agency’s] expertise”; therefore it “should be presumed immune from review under § 701(a)(2).” Id. at 830-32, 105 S.Ct. at 1655-56. The Chaney Court, however, also “emphasize[d] that the [nonenforcement] decision is only presumptively unreviewable; the presumption may be rebutted where the substantive statute has provided guidelines for the agency to follow in exercising its enforcement powers.” Id. at 832-33, 105 [803]*803S.Ct. at 1656. It is critical for our purposes to note that Chaney was explicitly based on § 701(a)(2) — the “no law to apply” section of the APA — and in no way implicated § 701(a)(1), the congressional preclusion exception.
A. The “Law to Apply” in This Case
The Chaney Court said it was “leaving to one side the problem of whether an agency’s rules might under certain circumstances provide courts with adequate guidelines for informed judicial review of decisions not to enforce.” Id. at 836, 105 S.Ct. at 1658. This case, however, squarely presents that situation in which an agency’s own regulations do contain a “judicially manageable” standard for making non-enforcement decisions. The “reasonable possibility” standard of NHTSA’s own regulations clearly requires NHTSA to make a factual judgment about the chances that a safety-related defect exists, based on the evidence compiled during the “technical review” prescribed by 49 C.F.R. § 552.6. This kind of factual judgment is definitely susceptible to judicial review, although, of course, under the APA, a reviewing court would set aside a NHTSA decision that no “reasonable possibility” of a safety-related defect existed only if there was no substantial support for that finding in the evidence before the Administrator. See Lorion v. Nuclear Regulatory Commission, 785 F.2d 1038 (D.C.Cir.1986); Association of Data Processing Service Organizations, Inc. (“ADPSCO”) v. Board of Governors, 745 F.2d 677 (D.C.Cir.1984). But although the court must accord the agency the appropriate degree of deference on such a record-based finding involving scientific and technological data, see Lorion, 785 F.2d at 1042, the court is still entirely capable of judging whether NHTSA’s “reasonable possibility” decision was “arbitrary” or “capricious.” 5 U.S.C. § 706(2)(A). Chaney’s concerns are therefore met.
Indeed, by confining the grounds for a decision to grant or deny a petition to the likelihood of a safety-related defect, § 552.8 itself rules out consideration of other discretionary factors like “whether agency resources are best spent on this violation or another” or “whether the particular enforcement action requested best fits the agency’s overall policies.” These other resource and priority factors, not at issue in NHTSA’s petition decisions, are the ones that Chaney’s rationale said rendered nonenforcement decisions presumptively unreviewable under § 701(a)(2). 2. Thus, the “reasonable possibility” standard NHTSA laid down in its own regulations amply rebuts the Chaney presumption against reviewability.
Indeed, if the “reasonable possibility” standard had appeared in the Motor Vehicle Safety Act itself, there could be no doubt that it provided a “judicially manageable” review standard. But the Supreme Court has told us that “[s]o long as [an administrative] regulation is extant it has the force of law,” United States v. Nixon, 418 U.S. 683, 695, 94 S.Ct. 3090, 3101, 41 L.Ed.2d 1039 (1974). Therefore, NHTSA’s own regulation containing the “reasonable possibility” review standard is the legal equivalent of a statutory standard for Chaney purposes.7 It is binding law governing the agency’s decisions in this realm and, as long as it is on the books, it must be followed. In sum, § 552.8 provides a reviewing court with “law to apply” for the purposes of § 701(a)(2). See California Human Development Corp. v. Brock, 762 F.2d 1044, 1048 n. 28 (D.C.Cir.1985); accord id. at 1053 (Scalia, J., concurring).8
[804]*804B. The Issue of Statutory Preclusion of Judicial Review
NHTSA, however, raises an additional objection to review, not ruled on by the District Court. It argues that the Motor Vehicle Safety Act has “affirmatively precluded” judicial review in this case.
In addressing this issue, we stress at the outset that there is nothing in the Act remotely resembling an express preclusion of judicial review of decisions not to grant citizen petitions. Although it is of course always possible in compelling circumstances to infer from the structure or history of a silent statute a congressional intent to preclude review, see Block v. Community Nutrition Institute, 467 U.S. 340, 104 S.Ct. 2450, 81 L.Ed.2d 270 (1984), that is clearly the unusual case, not the norm, and the evidence, we are told, must be very “clear and convincing” to overcome the normal presumption of review. See, e.g., Bowen v. Michigan Academy of Family Physicians, 476 U.S. 667, 106 S.Ct. 2133, 2136, 90 L.Ed.2d 623 (1986) (internal quotations omitted).
The government here relies on a few isolated and decidedly equivocal pieces of legislative history to make its case of affirmative preclusion. The House bill as originally introduced provided that “[i]f the Secretary denies the petition made under this section, ... the petitioner may commence a civil action in a United States district court to compel the Secretary to commence or complete the proceeding (or both), as requested in the petition,” H.R. 5529, § 7, 93d Cong., 1st Sess. (1973). The House Committee on Interstate and Foreign Commerce deleted this provision when it reported the bill to the floor. See H.R. Rep. No. 1191, 93d Cong., 2d Sess. 9 (1974), U.S.Code Cong. & Admin.News 1974 p. 6046. This skeletal sequence, however, fails, by itself, to establish an intent to preclude the normal avenues of judicial review open under the APA for agency decisions. Critically, it omits mentioning the crucial distinction between the kind of de novo, full-blown judicial review provided in the original House bill and the more limited judicial review regularly available under the APA, 5 U.S.C. § 706, which we are applying in this case.
The original version of H.R. 5529 stated that “if the petitioner can demonstrate to the satisfaction of the court, by a preponderance of the evidence in a de novo proceeding before such court, that the motor vehicle or motor vehicle equipment involved presents an unreasonable risk of injury ... or defect which relates to motor vehicle safety ... and that the failure of the Secretary to commence or complete the proceeding as requested in the petition unreasonably exposes the petitioner or other consumers to a risk of injury ... the court shall order the Secretary to commence or complete the proceedings (or both), as required in the petition.” H.R. Rep. 5529, § 7, 93d Cong., 1st Sess. (1973). A grant of judicial authority to make a de novo determination of safety risks when an agency has declined to undertake an enforcement action would indeed be a drastic innovation in the normal relationship between agencies and courts. By comparison, the usual, far more limited kind of judicial review under the APA involved here represents the norm for agency review, and to preclude it would be a distinct departure from Congress’ usual practice.
The APA prohibits a reviewing court from considering new evidence that was not before the agency when it made its decision. The APA does not permit a reviewing court to decide which side of the factual dispute it thinks “the preponderance of the evidence” in the administrative record falls on. Rather, as we have al[805]*805ready mentioned, judicial review under the APA requires that the agency’s decision be upheld as long as there is enough support in the record to assure it is not “arbitrary.” See ADPSO, supra, 745 F.2d at 683-84. Given these important differences between the usual form of judicial review accorded agency decisions and the House bill’s standard, one can well understand why Congress decided to delete the provision for de novo judicial review. It by no means logically follows, however, that by so doing Congress intended to preclude the much more limited form of judicial review generally available under the APA as well.
There is another crucial distinction between the omnibus type of judicial review Congress declined to provide here and the garden-type variety kind of judicial review we find applicable in this case. The judicial review provision of the original bill bound the Secretary to a specific legal standard for judging § 1410a petitions: whether or not “the motor vehicle equipment involved presents an unreasonable risk of injury.” H.R. 5529. In eliminating the judicial review provision, Congress was undoubtedly motivated in great part by its intention that the statute itself not subject the Secretary to a specific legal standard for- judging these petitions.
In this case, as we have seen, appellants are seeking review, not on the basis of the statute alone, but on the basis of a legal standard contained in the agency’s own regulations implementing the statute. Even when a statute grants an agency broad discretion in making a decision and itself provides no basis for review of that decision, it is well-settled that judicial review still exists to require the agency to follow procedural or substantive standards contained in its own regulations, which curtail the discretion conferred by statute. See, e.g., Service v. Dulles, 354 U.S. 363, 77 S.Ct. 1152, 1 L.Ed.2d 1403 (1957); Padula v. Webster, 822 F.2d 97 (D.C.Cir.1987). While we agree with the dissent that the existence of an administrative regulation cannot confer judicial review over an agency decision that Congress has affirmatively decided to render immune from judicial review, see Harrison v. Bowen, 815 F.2d 1505, 1517 (D.C.Cir.1987), that obvious proposition is not at issue here. Congress has in no way affirmatively precluded judicial review over the denial of enforcement petitions. Rather, the relevant dispute between ourselves and the dissent is whether this court should infer a congressional intent to preclude judicial review of an agency’s decision in a situation where review is sought on the basis of a legal standard contained in one of the agency’s own regulations. Unlike the dissent, we cannot conclude that by taking out the de novo judicial review provision of the original bill the House Committee — or the Congress as a whole — intended to prohibit a court from reviewing in an ordinary fashion under the APA a denial of a § 1410a petition in order to determine whether the agency adhered to a legal standard contained in one of its own regulations. Nor do we find any other evidence in the legislative history that Congress intended to preclude this limited form of judicial review.9
[806]*806C. The Service Principle and Its Applicability to This Case
The dissent argues a congressional intent to preclude judicial review of decisions denying § 1410a petitions primarily on the basis of the discretion granted the Secretary in § 1410a(c) as to how to handle these petitions. Such an inference, however, runs entirely counter to the principle that a court will require an agency to follow the legal standards contained in its own regulations despite the fact that a statute has granted the agency discretion in the matter. In Service v. Dulles, supra, the relevant statute granted the Secretary of State “absolute discretion” in deciding whether to terminate the employment of Foreign Service officers. 354 U.S. at 370, 77 S.Ct. at 1156 (quoting 60 Stat. 458). The Supreme Court nevertheless held that judicial review exists to determine whether the Secretary followed legal standards contained in his own regulations:
While it is of course true that under [the relevant statute] the Secretary was not obliged to impose upon himself these more rigorous substantive and procedural standards, neither was he prohibited from doing so, ... and having done so he could not, so long as the Regulations remained unchanged, proceed without regard to them.
Id. at 388, 77 S.Ct. at 1165. If a statutory grant of broad discretion provided a sufficient basis for inferring a congressional intent to render an agency’s decision immune from judicial review, even for the limited purposes of determining whether the agency complied with legal standards contained in its own regulations, the Supreme Court in Service would never, as it surely did, have instructed a court to conduct precisely that kind of review. But as the Service decision makes plain, a statutory grant of discretion does not provide a basis for inferring that Congress intended to prohibit a court from reviewing an administrative decision in order to determine whether the agency has adhered to its own legally binding standards. Rather, there must be a much more definitive source of evidence that Congress intended to immunize an administrative decision form of judicial review.
The dissent points us to no such source. It cites only the structure and legislative history of the Act as grounds for inferring an intent to preclude review in this case. But neither of these sources reveals any “fairly discernible” intent to preclude the form of review sought in this case, namely a determination of whether NHTSA adhered to the standard set forth in its own rule. Cf. Block v. Community Nutrition Institute, 467 U.S. 340, 104 S.Ct. 2450, 81 L.Ed.2d 270 (1984) (a case in which the statutory scheme of the Agricultural Marketing Agreement Act of 1937 presented a “fairly discernible” congressional intent to allow milk “handlers” but not milk consumers to challenge certain administrative pricing orders as unlawful under the Act). Arguing from structure, the dissent states that because the Act expressly provides for some form of judicial review at other stages of the administrative process, Congress must have intended that denials of § 1410a petitions be immune from review. But this questionable methodology for presuming an affirmative intent to preclude normal judicial review, even if appropriate for interpreting some other statutory scheme, is simply not legitimate with respect to the 1974 amendments to the Motor Vehicle Safety Act. For the Conference Report on the 1974 amendments states: “The conferees discussed but decided to take no position on whether or not pre-enforcement judicial review is available under the conference substitute. The conferees decided to leave that question to the courts.” Conf.Rep. 1452, 93d Cong., 2d Sess. 32 (1974). The dissent attempts to confine the impact of that statement to pre-enforcement actions by car manufacturers. There is no evidence of such a limited impact. Nonetheless, even if we accepted the dissent’s assumption that the term “pre-enforcement judicial review” in [807]*807this context referred only to circumstances in which a car manufacturer (rather than a car owner) seeks court review of a NHTSA decision prior to the time when the agency brings a civil enforcement action, it would still demonstrate our main point: the mere absence of an express provision of judicial review in the text of the statute for a particular kind of administrative decision cannot by itself mean that Congress intended to affirmatively preclude judicial review over that decision. Otherwise, there would be no point at all to the Conference Committee’s declination to decide on judicial review of any pre-enforcement decisions.10 Given this additional bolster in the legislative history that Congress did not intend to preclude judicial review in all situations where it did not affirmatively grant it, we are unwilling to infer an intent to prohibit review under the APA of a denial of a § 1410a petition to determine whether NHTSA complied with its own standards for denying such petitions.11
The legislative materials on which the dissent relies just do not add up to a “fairly discernible” intent to preclude the kind of judicial review sought in this case. The dissent’s structural argument is negated by specific legislative history showing that Congress definitely did not “carefully provide judicial review in specified circumstances but deliberately with [o]ld review” in other, “pre-enforcement” circumstances. Diss. op., infra at 817. The deletion of de novo review from the original House bill cannot be fairly taken as precluding judicial review under the deferential “arbitrary” or “capricious” standard of the APA when the [808]*808agency’s own regulations provide judicially manageable standards for that review. The only other basis on which the dissent relies for arguing preclusion of all review is the wide discretion the statute confers upon the Secretary in making these decisions. But, as we have discussed, reliance on this discretion to preclude review, when the statute itself is silent on the subject, and the legislative history is at best inconclusive, contradicts the principle of the Service case, which is that a court will require an agency to adhere to its own regulations “even when the administrative action under review is discretionary in nature.” Service, 354 U.S. at 372, 77 S.Ct. at 1157.
Nor do we believe that the Service principle no longer holds true because today’s case, unlike Service, involves a nonenforcement decision. The principle has been applied even in the context of criminal prosecutions. See United States v. Nixon, 418 U.S. 683, 695-96, 94 S.Ct. 3090, 3100-01, 41 L.Ed.2d 1039 (1974).12 Logically, there is no reason not to apply it to agency discretion concerning enforcement proceedings as well as to other forms of agency discretion.13 The Chaney presumption against judicial review of an agency’s nonenforcement decisions, derived from subsection (a)(2) of 5 U.S.C. § 701, is an adequate vehicle by which to prevent undue judicial interference into agency priority-resource decisions. That presumption, of course, stems from the usual lack' of “judicial manageable standards” governing these kinds of decisions and is rebuttable by the [809]*809existence of such a standard in a particular case.
In concluding this discussion, we emphasize that under the APA, there is no presumption against judicial review of an agency’s nonenforcement decision other than the Chaney presumption. Congress need not affirmatively signal the availability of judicial review when it enacts a statute concerning grants and denials of petitions for enforcement, like those involved in this case. Rather, as we have discussed, and Chaney makes perfectly clear, under subsection (a)(1) of § 701, the presumption works the other way. Chaney explicitly recognizes that earlier Supreme Court precedent construing § 701 “clearly separates the exception provided by § (a)(1) from the § (a)(2) exception.” 470 U.S. at 830, 105 S.Ct. at 1655. Thus, the fact that nonenforcement decisions are generally committed to agency discretion does not make them presumptively unreviewable under the § (a)(1) exception.
The dissent, however, acknowledges it would likely strike off in quite the opposite direction. It would presume nonenforcement decisions immune from judicial review unless Congress demonstrated affirmatively its intent to make these decisions reviewable. The dissent attempts to find some support for this position in the Chaney opinion, but in fact its view contradicts that opinion. Chaney states that if a statute circumscribes an agency’s enforcement discretion with legal standards, then the presumption against judicial review under § 701(a)(2) is rebutted. Chaney emphatically does not require additional evidence of a congressional intent to provide for judicial review of a nonenforcement decision under the APA where standards exist to govern the petition decisions. Thus, if the “reasonable possibility” standard of § 552.8 had been found in the statute itself, there could be no doubt that judicial review was available under the APA to determine whether NHTSA adhered to that standard. Post-Chaney precedents of this court require the same result where the standards are made by the agency itself. See Padula, supra; California Human Development Corp., supra.
Given Service and the multitude of cases following it, we take the position that, in the absence of any congressional intent to preclude review, judicial review is available to hold an agency to procedural and substantive standards contained in its own regulations governing an administrative decision, even when a statute grants the agency “absolute discretion” or something akin to it over that administrative decision. We do not understand Chaney as casting doubt on the validity of this longstanding presumption. The fact that non-enforcement decisions have been “traditionally” committed to agency discretion, Chaney, 470 U.S. at 832,105 S.Ct. at 1656, does not negate the presumption that when an agency by regulation circumscribes the discretion granted to it by Congress, judicial review is available to hold the agency to its rules.
This proposition represents the critical difference between the panel and the dissent’s argument in this case. We hold that when a statute grants an agency discretion but does not in text or by reasonable inference from legislative history or structure affirmatively preclude judicial review, review may exist under the Service principle. If, in that review, an agency’s own regulations provide a court with “law to apply,” then review may be had on the basis of compliance with that law. It is on such a basis that we find judicial review available in this case.
III. The Scope of Judicial Review in This Case
Because this case involves “informal agency action,” Florida Power & Light Co. v. Lorion, 470 U.S. 729, 744, 105 S.Ct. 1598, 1607, 84 L.Ed.2d 643 (1985), judicial review of NHTSA’s denial of appellants’ petition under the APA is limited to whether the decision was “arbitrary, capricious, or abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2)(A); Camp v. Pitts, 411 U.S. 138, 142, 93 S.Ct. 1241, 1244, 36 L.Ed.2d 106 (1973) (per curiam). Judicial review under § 706(2)(A) determines whether the reasons [810]*810for the agency’s decision were legally permissible and reasoned ones, and whether there was adequate factual support for the decision. See ADPSO, 745 F.2d at 683-84. The reviewing court may not consider new evidence that was not before the agency when it made its decision. See, e.g., Lorion, 470 U.S. at 744, 105 S.Ct. at 1607. “When the arbitrary and capricious standard is performing that function of assuring factual support, ... whether the administrator was arbitrary must be determined on the basis of what he had before him and not on the basis of ‘some new record made initially in the reviewing court.’ ” ADPSO, 745 F.2d at 683-84 (quoting Camp, 411 U.S. at 142, 93 S.Ct. at 1244).
To assure that the necessary factual support for the finding exists, the reviewing court normally must examine the evidence in the existing administrative record, even though it does so with deference to the agency’s judgment. “The APA specifically contemplates judicial review on the basis of the agency record compiled in the course of informal agency action in which a hearing has not occurred.” Lorion, 470 U.S. at 744, 105 S.Ct. at 1607.
The Supreme Court has, however, in Dunlop v. Bachowski, 421 U.S. 560, 95 S.Ct. 1851, 44 L.Ed.2d 377 (1975), ruled that in some exceptional circumstances it may be improper for a court to go behind the agency’s facial rationale and look into the factual basis for its decision. Dunlop involved a decision by the Secretary of Labor not to bring suit to set aside a union election in violation of the Labor-Management Reporting and Disclosure Act of 1959 (LMRDA). The Court held that while the LMRDA did not entirely prohibit judicial review of the Secretary’s decision, it did ban a judicial “challenge to the factual basis for the Secretary’s decision.” 421 U.S. at 577, 95 S.Ct. at 1862; see id. at 573, 95 S.Ct. at 1860. Thus, the Supreme Court stated, “the court’s review should be confined to examination of the [Secretary’s] ‘reasons’ statement, and the determination whether the statement, without more, evinces that the Secretary’s decision is so irrational as to constitute the decision arbitrary and capricious.” Id. at 572-73, 95 S.Ct. at 1860. Dunlop limited the APA’s “arbitrary or capricious” standard in that case “to determin[ing] whether the Secretary’s decision was reached for an impermissible reason or no reason at all,” id. at 573, 95 S.Ct. at 1861, thereby eliminating the second function of normal APA review, assuring that factual support for the decision existed.
The Dunlop Court, in limiting review, said it was basing its decision on the “congressional objectives” underlying the LMRDA. According to the Dunlop opinion, “[f]our prior decisions of the Court construing the LMRDA identified] the congressional objectives ... not to permit individuals to block or delay resolution of post-election disputes, but rather ‘to settle as quickly as practicable the cloud on incumbents’ titles to office;’ and ‘to protect unions from frivolous litigation and unnecessary interference with their elections.' ” 421 U.S. at 568, 573, 95 S.Ct. at 1858, 1860 (internal quotation omitted). The Supreme Court reasoned that allowing court “challenges to the factual basis for the Secretary’s conclusion either that no violations occurred or that they did not affect the outcome of the electionfs] ... would be defiant of [these] congressional objectives.” Id. at 573, 95 S.Ct. at 1860.
The Court’s specifically articulated concerns are, of course, unique to the LMRDA context, and obviously not applicable in the Motor Vehicle Safety Act context. Some opinions in and outside this circuit have accordingly refused to apply the Dunlop exception to normal APA review in circumstances outside the context of union election disputes under the LMRDA.14 For [811]*811example, in NRDC v. SEC, 606 F.2d 1031 (D.C.Cir.1979), we rejected the government’s argument, based on Dunlop, that judicial review of the SEC’s decision not to promulgate rules requiring corporations to disclose their environmental policies was limited to the SEC’s statement of reasons for its decision. Id. at 1053 n. 31. Instead, we held that the SEC’s decision was subject to the “usual rule” that judicial review under the “arbitrary” or “capricious” standard requires the court to examine the existing administrative record to assure that the agency had factual support for its decision. Id. Similarly, the Third Circuit, in an opinion authored by Judge Arlin Adams, held that the limited Dunlop form of judicial review did not apply to decisions of the Environmental Protection Agency involving the sulphur oxide emissions of certain utility companies. See Duquesne Light Co. v. EPA, 522 F.2d 1186,1193 n. 24 (3d Cir.1975). The court then proceeded to examine the administrative record in the case to determine whether the EPA had the necessary factual support. See id. at 1196.
Other opinions, however, again from both this and other circuits, have applied Dunlop outside the LMRDA context. In Exxon Pipeline Co. v. United States, 725 F.2d 1467 (D.C.Cir.1984), we limited judicial review of decisions by the Federal Energy Regulatory Commission (FERC) to suspend certain pipeline rates to an examination of the agency’s stated reasons for the suspension. In Ward v. Parole Commission, 804 F.2d 64, 67 (7th Cir.1986), the Seventh Circuit applied Dunlop to a decision by the federal Bureau of Prisons, transferring an inmate from one prison to another. See also Kitchens v. Department of Treasury, 535 F.2d 1197 (9th Cir.1976) (applying Dun-lop to a decision of the Bureau of Alcohol, Tobacco, and Firearms under the Gun Control Act of 1968).
Like the Dunlop opinion itself, none of these circuit court opinions set forth a precise standard for determining when the arbitrary or capricious standard performs its usual function of factual support and when, by contrast, judicial review is confined to an examination of the agency’s statement of reasons. The' dissent reads the caselaw as holding that in order to decide whether the Dunlop exception applies, we must look to see whether a congressional statute grants an administrative agency “special” discretion to make a certain decision. Diss. op., infra at 822.15 We must confess our skepticism as to how much, if anything, a “special discretion” formula adds to the inquiry. Certainly we agree the statutory grant of discretion must be “special,” in some sense, to invoke Dun-lop’s limited review standard since the simple fact that a statute grants an agency “discretion” cannot mean that judicial re[812]*812view is limited to the agency’s statement of reasons. If that were the case, the Dun-lop exception would become the “Jaws” of administrative law, gobbling up the usual APA rule that judicial review, even of informal, discretionary decisions, is based on the administrative record (as long as there is “law to apply”). See Lorion, 470 U.S. at 744, 105 S.Ct. at 1607.
But what exactly does it mean for an agency’s discretion under a statute to be “special”? The dissent fails to enlighten us in this regard. For ourselves, we think the only pertinent inquiry is whether there is evidence from the structure or history of the law that Congress meant to exempt the agency’s decision from the usual modus operandi of judicial review under 5 U.S.C. § 706(2)(A).16 We do not see that looking for “special” discretion in a statute amounts to anything different than looking for a congressional intent to limit judicial review.17 That intent, as in Dunlop, may, of course, be gleaned from specific congressional objectives that conflict with ordinary review.
Nor do we agree with what we understand to be the implication of the dissent’s reasoning, that even if the Chaney presumption of unreviewability of an agency’s nonenforcement decision is rebutted, review of the nonenforcement decision must automatically be limited to the agency’s statement of reasons.18 Neither Dun-lop nor Chaney ever stated so broad a holding, and the caselaw even after Chaney supports review of nonenforcement decisions for factual support. For example, [813]*813in the Lorion case, this court reviewed a decision of the Nuclear Regulatory Commission (NRC) denying a petition to commence an enforcement proceeding against an allegedly unsafe nuclear reactor to determine whether the NRC had any factual support for its determination that the petition did not present “a substantial [safety] issue.” 785 F.2d at 1042. In our view, Dunlop stands for the considerably narrower principle that courts will not conduct their usual review of the evidence in the administrative record underlying the agency’s decision if this kind of review would interfere with identifiable congressional objectives of the relevant statute.
Here, we can identify no congressional objectives of the Motor Vehicle Safety Act with which the normal level of judicial review would interfere. The basic congressional objective underlying the Motor Vehicle Safety Act is stated simply and clearly in the Act’s first section: “Congress hereby declares that the purpose of this chapter is to reduce traffic accidents and deaths and injuries to persons resulting from traffic accidents.” Pub.L. No. 89-563, § 1, 80 Stat. 718 (codified as 15 U.S.C. § 1381). To the extent that judicial review helps to assure that factual support exists for NHTSA decisions denying enforcement petitions, it helps to reduce the threat of traffic accidents, and aids, not hinders, the basic congressional purpose of the statute.
Moreover, the Motor Vehicle Safety Act expressly contemplates extensive citizen involvement in enforcement of the Act by providing for citizen petitions for enforcement of safety standards, by expressly authorizing public hearings on allegations of safety-related defects, and by requiring the agency to publish a statement of reasons for denying a petition. This statutory scheme starkly contrasts with the one involved in Dunlop. The LMRDA did not itself require any statement of reasons for the Secretary’s decision to set aside elections based on union misconduct, and it certainly did not contain any provision authorizing the Secretary of Labor to conduct a public hearing on allegations of misconduct involving union elections, at which “any interested person” could participate and introduce evidence. Such provisions would have been totally inconsistent with the congressional objective “not to permit individuals to block or delay resolution of post-election disputes.” The explicit provisions for citizen participation contained in the motor safety enforcement provisions of 15 U.S.C. § 1410a demonstrate that there were no Dunlop-type congressional objectives in the Act with which we are concerned.
Also, the nonenforcement decision under review here differs from that involved in Dunlop in a way that strongly suggests the inapplicability of that precedent to this case. A decision concerning allegations of misconduct in a union election requires a judgment about the actions and intentions of particular individuals. It would be virtually impossible for a reviewing court to revisit those kinds of factual decisions without conducting an independent “trial-type” inquiry or impermissibly substituting its judgment for the agency. The Dunlop Court was at great pains to avoid this kind of “trial-type” inquiry. The Court twice referred to “[t]he full trappings” of adversary trial-type hearings as the kind of judicial review which would interfere with the congressional objectives underlying the LMRDA. 421 U.S. at 573, 577, 95 S.Ct. at 1860, 1862.
By contrast, the administrative decision in this case concerns the possibility that certain kinds of cars are defective and unsafe. This possibility, in turn, depends on an evaluation of “technical” evidence about the cars which the agency’s own regulations prescribe. See 49 C.F.R. § 552.6. Of course, the reviewing court may not substitute its judgment for the agency’s but it certainly is possible for the court to review the administrative decision for factual support without an impermissible substitution of judgment, in the same way we review agency factual determinations under § 706(2)(A) thousands of times every year. Cf. Motor Vehicle Manufacturers Association v. State Farm Mutual Automobile Insurance, 463 U.S. 29, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983) (holding that the arbitrary and capricious standard performs its [814]*814usual role of assuring factual support in a case involving a NHTSA decision to rescind a motor «vehicle safety standard).
Thus, the concerns about identifiable congressional objectives that led the Dunlop court to limit judicial review in that case simply are not present here. There is no reason to bypass the normal function of the “arbitrary” or “capricious” standard of review, i.e., to assure that factual support for the agency’s decision exists. As we noted above, the Lorion case, remarkably similar in context to this case, stands as a firm precedent for factual review under § 706(2)(A).
Lorion involved an NRC decision denying a citizen petition for enforcement. Under its own rules, the NRC is required to grant the petition if the evidence presents a “substantial ... safety issue” concerning the design of a nuclear reactor. See Lorion, 470 U.S. at 732, 105 S.Ct. at 1601. Lorion, 785 F.2d at 1042. This administrative standard corresponds to the “reasonable possibility” standard in NHTSA’s regulations. Thus, in both cases, the agency’s decision to grant or deny the citizen petition for enforcement turns on whether the evidence in the administrative record demonstrates a significant threat to public safety.
The Lorion case produced opinions from the Supreme Court and our own circuit. The question presented to the Supreme Court on writ of certiorari was whether jurisdiction to review the NRC decision resided exclusively in the court of appeals, or in the district court as well. But, in order to decide this question, the Supreme Court had to consider the nature of the judicial review that it expected would take place.19 In so doing, the Court made very clear that judicial review would be based on the existing administrative record:
As the actions of the Commission in compiling a 547-page record in this case demonstrate, agencies typically compile records in the course of informal agency action. The APA specifically contemplates judicial review on the basis of the administrative record compiled in the course of informal agency action in which a hearing has not occurred.
470 U.S. at 744, 105 S.Ct. at 1607.
And indeed this court on remand performed precisely the kind of judicial review expected by the Supreme Court opinion. Our review was appropriately deferential, but we examined the administrative record, referred to by the Supreme Court, to determine whether it supported the NRC’s determination that the citizen’s petition presented no “substantial safety issue.” 785 F.2d at 1043. Together, the Supreme Court opinion in Lorion and the subsequent Lorion decision in our own circuit, establish a strong precedent in favor of normal APA record review in this case.
Both Lorion and this case involve public safety issues of great import. One concerns the design of a nuclear reactor; the other, the design of an automotive transmission. But the basic factual issue up for review is the same: does the product’s design in fact contain a defect that may cause widespread injuries, and perhaps fatalities? Indeed, the public safety issues in both cases were deemed sufficiently important by both agencies as to require in their own regulations that safety be the sole consideration in deciding enforcement petitions.20 While this court cannot, and should not, second-guess the agency’s safety determinations, this court can and should assure that the agency has some factual support for those safety determinations. Making sure that the agency’s stated reason for denying an enforcement petition is in truth the lack of a “substantial safety issue” guarantees that the responsible administrative agency has given at least a minimum amount of consideration to the citizens whose safety Congress has entrusted to the agency. In contrast to Dunlop, confining judicial review to the agency’s statement of reasons would em[815]*815phatically not serve basic congressional objectives in this area of the law.
Our holding then is this: When Congress grants citizens the right to petition administrative agencies for enforcement on an issue of public safety, and the relevant substantive statute sends out no identifiable signals on the availability or scope of judicial review of an agency’s denial of that petition, but the agency has bound itself by a legal rule to grant the petition unless it makes a finding of “no reasonable possibility” of a safety defect, then a petitioner is entitled under the APA to judicial review of a decision denying the petition to assure that the agency had some factual support for its “no reasonable possibility” finding. In such circumstances, the arbitrary or capricious standard of review in § 706 of the APA functions as usual, and is not limited to examining merely the stated reasons for the agency’s decision. Therefore, following the Lorion precedent in this circuit, we hold that NHTSA’s determination that no “reasonable possibility” of a safety-related defect exists in this case is subject to judicial review on the basis of the evidence that the Administrator had before her when she made her decision.
Conclusion
The Motor Vehicle Safety Act does not “affirmatively preclude” judicial review of NHTSA decisions denying citizen petitions for enforcement against alleged automotive safety defects. NHTSA is bound by its own regulation requiring that it grant a citizen petition if there is a “reasonable possibility” that a safety-related defect exists. This “reasonable possibility” standard provides the court with a judicially manageable standard for reviewing NHTSA petition decisions, thereby rebutting the presumption against judicial review of nonenforcement decisions, established in Chaney.
The scope of judicial review in this case falls under the “arbitrary” or “capricious” ■ standard of 5 U.S.C. § 706(2)(A), which in this case requires both that the agency’s stated reason for denying a citizen petition conform to the rule set forth in its own regulations and that the agency have a sufficient factual basis for its decision. The reviewing court should be appropriately deferential, as mandated by our Lorion opinion. Nevertheless, the reviewing court must examine the record on which the agency relied. Because the District Court did not examine the administrative record in this case, we reverse its decision and remand for an appropriate review.
Reversed and remanded.