Justice White
delivered the opinion of the Court.
The development of the automobile gave Americans unprecedented freedom to travel, but exacted a high price for [33]*33enhanced mobility. Since 1929, motor vehicles have been the leading cause of accidental deaths and injuries in the United States. In 1982, 46,300 Americans died in motor vehicle accidents and hundreds of thousands more were maimed and injured.1 While a consensus exists that the current loss of life on our highways is unacceptably high, improving safety does not admit to easy solution. In 1966, Congress decided that at least part of the answer lies in improving the design and safety features of the vehicle itself.2 But much of the technology for building safer cars was undeveloped or untested. Before changes in automobile design could be mandated, the effectiveness of these changes had to be studied, their costs examined, and public acceptance considered. This task called for considerable expertise and Congress responded by enacting the National Traffic and Motor Vehicle Safety Act of 1966 (Act), 80 Stat. 718, as amended, 15 U. S. C. § 1381 et seq. (1976 ed. and Supp. V). The Act, created for the purpose of “reducing] traffic accidents and deaths and injuries to persons resulting from traffic accidents,” 15 U. S. C. § 1381, directs the Secretary of Transportation or his delegate to issue motor vehicle safety standards that “shall be practicable, shall meet the need for motor vehicle safety, and shall be stated in objective terms.” 15 U. S. C. § 1392(a) (1976 ed., Supp. V). In issuing these standards, the Secretary is directed to consider “relevant available motor vehicle safety data,” whether the proposed standard “is reasonable, practicable and appropriate” for the particular type of motor vehicle, and the “extent to which [34]*34such standards will contribute to carrying out the purposes” of the Act. 15 U. S. C. §§ 1392(f)(1), (3), (4).3
The Act also authorizes judicial review under the provisions of the Administrative Procedure Act (APA), 5 U. S. C. §706, of all “orders establishing, amending, or revoking a Federal motor vehicle safety standard,” 15 U. S. C. § 1392(b). Under this authority, we review today whether NHTSA acted arbitrarily and capriciously in revoking the requirement in Motor Vehicle Safety Standard 208 that new motor vehicles produced after September 1982 be equipped with passive restraints to protect the safety of the occupants of the vehicle in the event of a collision. Briefly summarized, we hold that the agency failed to present an adequate basis and explanation for rescinding the passive restraint requirement and that the agency must either consider the matter further or adhere to or amend Standard 208 along lines which its analysis supports. •
I
The regulation whose rescission is at issue bears a complex and convoluted history. Over the course of approximately 60 rulemaking notices, the requirement has been imposed, amended, rescinded, reimposed, and now rescinded again.
As originally issued by the Department of Transportation in 1967, Standard 208 simply required the installation of seatbelts in all automobiles. 32 Fed. Reg. 2415. It soon became apparent that the level of seatbelt use was too low to reduce traffic injuries to an acceptable level. The Department therefore began consideration of “passive occupant restraint systems” — devices that do not depend for their effec[35]*35tiveness upon any action taken by the occupant except that necessary to operate the vehicle. Two types of automatic crash protection emerged: automatic seatbelts and airbags. The automatic seatbelt is a traditional safety belt, which when fastened to the interior of the door remains attached without impeding entry or exit from the vehicle, and deploys automatically without any action on the part of the passenger. The airbag is an inflatable device concealed in the dashboard and steering column. It automatically inflates when a sensor indicates that deceleration forces from an accident have exceeded a preset minimum, then rapidly deflates to dissipate those forces. The lifesaving potential of these devices was immediately recognized, and in 1977, after substantial on-the-road experience with both devices, it was estimated by NHTSA that passive restraints could prevent approximately 12,000 deaths and over 100,000 serious injuries annually. 42 Fed. Reg. 34298.
In 1969, the Department formally proposed a standard requiring the installation of passive restraints, 34 Fed. Reg. 11148, thereby commencing a lengthy series of proceedings. In 1970, the agency revised Standard 208 to include passive protection requirements, 35 Fed. Reg. 16927, and in 1972, the agency amended the Standard to require full passive protection for all front seat occupants of vehicles manufactured after August 15, 1975. 37 Fed. Reg. 3911. In the interim, vehicles built between August 1973 and August 1975 were to carry either passive restraints or lap and shoulder belts coupled with an “ignition interlock” that would prevent starting the vehicle if the belts were not connected.4 On review, the [36]*36agency’s decision to require passive restraints was found to be supported by “substantial evidence” and upheld. Chrysler Corp. v. Department of Transportation, 472 F. 2d 659 (CA6 1972).5
In preparing for the upcoming model year, most car makers chose the “ignition interlock” option, a decision which was highly unpopular, and led Congress to amend the Act to prohibit a motor vehicle safety standard from requiring or permitting compliance by means of an ignition interlock or a continuous buzzer designed to indicate that safety belts were not in use. Motor Vehicle and Schoolbus Safety Amendments of 1974, Pub. L. 93-492, §109, 88 Stat. 1482, 15 U. S. C. § 1410b(b). The 1974 Amendments also provided that any safety standard that could be satisfied by a system other than seatbelts would have to be submitted to Congress where it could be vetoed by concurrent resolution of both Houses. 15 U. S. C. § 1410b(b)(2).6
The effective date for mandatory passive restraint systems was extended for a year until August 31,1976. 40 Fed. Reg. 16217 (1975); id., at 33977. But in June 1976, Secretary of Transportation William T. Coleman, Jr., initiated a new rulemaking on the issue, 41 Fed. Reg. 24070. After hearing testimony and reviewing written comments, Coleman extended the optional alternatives indefinitely and suspended the passive restraint requirement. Although he found pas[37]*37sive restraints technologically and economically feasible, the Secretary based his decision on the expectation that there would be widespread public resistance to the new systems. He instead proposed a demonstration project involving up to 500,000 cars installed with passive restraints, in order to smooth the way for public acceptance of mandatory passive restraints at a later date. Department of Transportation, The Secretary’s Decision Concerning Motor Vehicle Occupant Crash Protection (Dec. 6, 1976), App. 2068.
Coleman’s successor as Secretary of Transportation disagreed. Within months of assuming office, Secretary Brock Adams decided that the demonstration project was unnecessary. He issued a new mandatory passive restraint regulation, known as Modified Standard 208. 42 Fed. Reg. 34289 (1977); 49 CFR § 571.208 (1978). The Modified Standard mandated the phasing in of passive restraints beginning with large cars in model year 1982 and extending to all cars by model year 1984. The two principal systems that would satisfy the Standard were airbags and passive belts; the choice of which system to install was left to the manufacturers. In Pacific Legal Foundation v. Department of Transportation, 193 U. S. App. D. C. 184, 593 F. 2d 1338, cert. denied, 444 U. S. 830 (1979), the Court of Appeals upheld Modified Standard 208 as a rational, nonarbitrary regulation consistent with the agency’s mandate under the Act. The Standard also survived scrutiny by Congress, which did not exercise its authority under the legislative veto provision of the 1974 Amendments.7
Over the next several years, the automobile industry geared up to comply with Modified Standard 208. As late as July 1980, NHTSA reported:
[38]*38“On the road experience in thousands of vehicles equipped with air bags and automatic safety belts has confirmed agency estimates of the life-saving and injury-preventing benefits of such systems. When all cars are equipped with automatic crash protection systems, each year an estimated 9,000 more lives will be saved, and tens of thousands of serious injuries will be prevented.” NHTSA, Automobile Occupant Crash Protection, Progress Report No. 3, p. 4; App. in No. 81-2220 (CADC), p. 1627 (hereinafter App.).
In February 1981, however, Secretary of Transportation Andrew Lewis reopened the rulemaking due to changed economic circumstances and, in particular, the difficulties of the automobile industry. 46 Fed. Reg. 12033. Two months later, the agency ordered a one-year delay in the application of the Standard to large cars, extending the deadline to September 1982, id., at 21172, and at the same time, proposed the possible rescission of the entire Standard. Id., at 21205. After receiving written comments and holding public hearings, NHTSA issued a final rule (Notice 25) that rescinded the passive restraint requirement contained in Modified Standard 208.
II
In a statement explaining the rescission, NHTSA maintained that it was no longer able to find, as it had in 1977, that the automatic restraint requirement would produce significant safety benefits. Notice 25, id., at 53419. This judgment reflected not a change of opinion on the effectiveness of the technology, but a change in plans by the automobile industry. In 1977, the agency had assumed that airbags would be installed in 60% of all new cars and automatic seatbelts in 40%. By 1981 it became apparent that automobile manufacturers planned to install the automatic seatbelts in approximately 99% of the new cars. For this reason, the lifesaving potential of airbags would not be realized. Moreover, it now appeared that the overwhelming majority of passive belts [39]*39planned to be installed by manufacturers could be detached easily and left that way permanently. Passive belts, once detached, then required “the same type of affirmative action that is the stumbling block to obtaining high usage levels of manual belts.” Id., at 53421. For this reason, the agency concluded that there was no longer a basis for reliably predicting that the Standard would lead to any significant increased usage of restraints at all.
In view of the possibly minimal safety benefits, the automatic restraint requirement no longer was reasonable or practicable in the agency’s view. The requirement would require approximately $1 billion to implement and the agency did not believe it would be reasonable to impose such substantial costs on manufacturers and consumers without more adequate assurance that sufficient safety benefits would accrue. In addition, NHTSA concluded that automatic restraints might have an adverse effect on the public’s attitude toward safety. Given the high expense and limited benefits of detachable belts, NHTSA feared that many consumers would regard the Standard as an instance of ineffective regulation, adversely affecting the public’s view of safety regulation and, in particular, “poisoning . . . popular sentiment toward efforts to improve occupant restraint systems in the future.” Id., at 53424.
State Farm Mutual Automobile Insurance Co. and the National Association of Independent Insurers filed petitions for review of NHTSA’s rescission of the passive restraint Standard. The United States Court of Appeals for the District of Columbia Circuit held that the agency’s rescission of the passive restraint requirement was arbitrary and capricious. 220 U. S. App. D. C. 170, 680 F. 2d 206 (1982). While observing that rescission is not unrelated to an agency’s refusal to take action in the first instance, the court concluded that, in this case, NHTSA’s discretion to rescind the passive restraint requirement had been restricted by various forms of congressional “reaction” to the passive restraint issue. It then [40]*40proceeded to find that the rescission of Standard 208 was arbitrary and capricious for three reasons. First, the court found insufficient as a basis for rescission NHTSA’s conclusion that it could not reliably predict an increase in belt usage under the Standard. The court held that there was insufficient evidence in the record to sustain NHTSA’s position on this issue, and that, “only a well justified refusal to seek more evidence could render rescission non-arbitrary.” Id., at 196, 680 F. 2d, at 232. Second, a majority of the panel8 concluded that NHTSA inadequately considered the possibility of requiring manufacturers to install nondetachable rather than detachable passive belts. Third, the majority found that the agency acted arbitrarily and capriciously by failing to give any consideration whatever to requiring compliance with Modified Standard 208 by the installation of airbags.
The court allowed NHTSA 30 days in which to submit a schedule for “resolving the questions raised in th[e] opinion.” Id., at 206, 680 F. 2d, at 242. Subsequently, the agency filed a Notice of Proposed Supplemental Rulemaking setting forth a schedule for complying with the court’s mandate. On August 4, 1982, the Court of Appeals issued an order staying the compliance date for the passive restraint requirement until September 1, 1983, and requested NHTSA to inform the court whether that compliance date was achievable. NHTSA informed the court on October 1,1982, that based on representations by manufacturers, it did not appear that practicable compliance could be achieved before September 1985. On November 8, 1982, we granted certiorari, 459 U. S. 987, and on November 18, the Court of Appeals entered an order recalling its mandate.
r-H h-I
Unlike the Court of Appeals, we do not find the appropriate scope of judicial review to be the “most troublesome [41]*41question” in these cases. Both the Act and the 1974 Amendments concerning occupant crash protection standards indicate that motor vehicle safety standards are to be promulgated under the informal rulemaking procedures of the Administrative Procedure Act. 5 U. S. C. § 553. The agency’s action in promulgating such standards therefore may be set aside if found to be “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 5 U. S. C. § 706(2)(A); Citizens to Preserve Overton Park v. Volpe, 401 U. S. 402, 414 (1971); Bowman Transportation, Inc. v. Arkansas-Best Freight System, Inc., 419 U. S. 281 (1974). We believe that the rescission or modification of an occupant-protection standard is subject to the same test. Section 103(b) of the Act, 15 U. S. C. § 1392(b), states that the procedural and judicial review provisions of the Administrative Procedure Act “shall apply to all orders establishing, amending, or revoking a Federal motor vehicle safety standard,” and suggests no difference in the scope of judicial review depending upon the nature of the agency’s action.
Petitioner Motor Vehicle Manufacturers Association (MVMA) disagrees, contending that the rescission of an agency rule should be judged by the same standard a court would use to judge an agency’s refusal to promulgate a rule in the first place — a standard petitioner believes considerably narrower than the traditional arbitrary-and-capricious test. We reject this view. The Act expressly equates orders “revoking” and “establishing” safety standards; neither that Act nor the APA suggests that revocations are to be treated as refusals to promulgate standards. Petitioner’s view would render meaningless Congress’ authorization for judicial review of orders revoking safety rules. Moreover, the revocation of an extant regulation is substantially different than a failure to act. Revocation constitutes a reversal of the agency’s former views as to the proper course. A “settled course of behavior embodies the agency’s informed judgment that, by pursuing that course, it will carry out the policies [42]*42committed to it by Congress. There is, then, at least a presumption that those policies will be carried out best if the settled rule is adhered to.” Atchison, T. & S. F. R. Co. v. Wichita Bd. of Trade, 412 U. S. 800, 807-808 (1973). Accordingly, an agency changing its course by rescinding a rule is obligated to supply a reasoned analysis for the change beyond that which may be required when an agency does not act in the first instance.
In so holding, we fully recognize that “[rjegulatory agencies do not establish rules of conduct to last forever,” American Trucking Assns., Inc. v. Atchison, T. & S. F. R. Co., 387 U. S. 397, 416 (1967), and that an agency must be given ample latitude to “adapt their rules and policies to the demands of changing circumstances.” Permian Basin Area Rate Cases, 390 U. S. 747, 784 (1968). But the forces of change do not always or necessarily point in the direction of deregulation. In the abstract, there is no more reason to presume that changing circumstances require the rescission of prior action, instead of a revision in or even the extension of current regulation. If Congress established a presumption from which judicial review should start, that presumption — contrary to petitioners’ views — is not against safety regulation, but against changes in current policy that are not justified by the rulemaking record. While the removal of a regulation may not entail the monetary expenditures and other costs of enacting a new standard, and, accordingly, it may be easier for an agency to justify a deregulatory action, the direction in which an agency chooses to move does not alter the standard of judicial review established by law.
The Department of Transportation accepts the applicability of the “arbitrary and capricious” standard. It argues that under this standard, a reviewing court may not set aside an agency rule that is rational, based on consideration of the relevant factors, and within the scope of the authority delegated to the agency by the statute. We do not disagree with [43]*43this formulation.9 The scope of review under the “arbitrary and capricious” standard is narrow and a court is not to substitute its judgment for that of the agency. Nevertheless, the agency must examine the relevant data and articulate a satisfactory explanation for its action including a “rational connection between the facts found and the choice made.” Burlington Truck Lines, Inc. v. United States, 371 U. S. 156, 168 (1962). In reviewing that explanation, we must “consider whether the decision was based on a consideration of the relevant factors and whether there has been a clear error of judgment.” Bowman Transportation, Inc. v. Arkansas-Best Freight System, Inc., supra, at 285; Citizens to Preserve Overton Park v. Volpe, supra, at 416. Normally, an agency rule would be arbitrary and capricious if the agency has relied on factors which Congress has not intended it to consider, entirely failed to consider an important aspect of the problem, offered an explanation for its decision that runs counter to the evidence before the agency, or is so implausible that it could not be ascribed to a difference in view or the product of agency expertise. The reviewing court should not attempt itself to make up for such deficiencies; we may not supply a reasoned basis for the agency’s action that the agency itself has not given. SEC v. Chenery Corp., 332 U. S. 194, 196 (1947). We will, however, “uphold a decision of less than ideal clarity if the agency’s path may reasonably be discerned.” Bowman Transportation, Inc. v. Arkansas-Best Freight System, Inc., supra, at 286. See also Camp v. Pitts, 411 U. S. 138, 142-143 (1973) (per curiam). For purposes of these cases, it is also relevant that Congress required a record of the rulemaking proceedings to be compiled [44]*44and submitted to a reviewing court, 15 U. S. C. § 1394, and intended that agency findings under the Act would be supported by “substantial evidence on the record considered as a whole.” S. Rep. No. 1301, 89th Cong., 2d Sess., 8 (1966); H. R. Rep. No. 1776, 89th Cong., 2d Sess., 21 (1966).
>
The Court of Appeals correctly found that the arbitrary- and-capricious test applied to rescissions of prior agency regulations, but then erred in intensifying the scope of its review based upon its reading of legislative events. It held that congressional reaction to various versions of Standard 208 “raise[d] doubts” that NHTSA’s rescission “necessarily demonstrates an effort to fulfill its statutory mandate,” and therefore the agency was obligated to provide “increasingly clear and convincing reasons” for its action. 220 U. S. App. D. C., at 186, 193, 680 F. 2d, at 222, 229. Specifically, the Court of Appeals found significance in three legislative occurrences:
“In 1974, Congress banned the ignition interlock but did not foreclose NHTSA’s pursuit of a passive restraint standard. In 1977, Congress allowed the standard to take effect when neither of the concurrent resolutions needed for disapproval was passed. In 1980, a majority of each house indicated support for the concept of mandatory passive restraints and a majority of each house supported the unprecedented attempt to require some installation of airbags.” Id., at 192, 680 F. 2d, at 228.
From these legislative acts and nonacts the Court of Appeals derived a “congressional commitment to the concept of automatic crash protection devices for vehicle occupants.” Ibid.
This path of analysis was misguided and the inferences it produced are questionable. It is noteworthy that in this Court respondent State Farm expressly agrees that the post-enactment legislative history of the Act does not heighten the [45]*45standard of review of NHTSA’s actions. Brief for Respondent State Farm Mutual Automobile Insurance Co. 13. State Farm’s concession is well taken for this Court has never suggested that the standard of review is enlarged or diminished by subsequent congressional action. While an agency’s interpretation of a statute may be confirmed or ratified by subsequent congressional failure to change that interpretation, Bob Jones University v. United States, 461 U. S. 574, 599-602 (1983); Haig v. Agee, 453 U. S. 280, 291-300 (1981), in the cases before us, even an unequivocal ratification — short of statutory incorporation — of the passive restraint standard would not connote approval or disapproval of an agency’s later decision to rescind the regulation. That decision remains subject to the arbitrary-and-capricious standard.
That we should not be so quick to infer a congressional mandate for passive restraints is confirmed by examining the postenactment legislative events cited by the Court of Appeals. Even were we inclined to rely on inchoate legislative action, the inferences to be drawn fail to suggest that NHTSA acted improperly in rescinding Standard 208. First, in 1974 a mandatory passive restraint standard was technically not in effect, see n. 6, supra; Congress had no reason to foreclose that course. Moreover, one can hardly infer support for a mandatory standard from Congress’ decision to provide that such a regulation would be subject to disapproval by resolutions of disapproval in both Houses. Similarly, no mandate can be divined from the tabling of resolutions of disapproval which were introduced in 1977. The failure of Congress to exercise its veto might reflect legislative deference to the agency’s expertise and does not indicate that Congress would disapprove of the agency’s action in 1981. And even if Congress favored the Standard in 1977, it — like NHTSA — may well reach a different judgment, given changed circumstances four years later. Finally, the Court of Appeals read too much into floor action on the 1980 authorization bill, a bill which was not enacted into law. Other [46]*46contemporaneous events could be read as showing equal congressional hostility to passive restraints.10
V
The ultimate question before us is whether NHTSA’s rescission of the passive restraint requirement of Standard 208 was arbitrary and capricious. We conclude, as did the Court of Appeals, that it was. We also conclude, but for somewhat different reasons, that further consideration of the issue by the agency is therefore required. We deal separately with the rescission as it applies to airbags and as it applies to seatbelts.
A
The first and most obvious reason for finding the rescission arbitrary and capricious is that NHTSA apparently gave no consideration whatever to modifying the Standard to require that airbag technology be utilized. Standard 208 sought to achieve automatic crash protection by requiring automobile manufacturers to install either of two passive restraint devices: airbags or automatic seatbelts. There was no suggestion in the long rulemaking process that led to Standard 208 that if only one of these options were feasible, no passive restraint standard should be promulgated. Indeed, the agency’s original proposed Standard contemplated the installation of inflatable restraints in all cars.11 Automatic belts [47]*47were added as a means of complying with the Standard because they were believed to be as effective as airbags in achieving the goal of occupant crash protection. 36 Fed. Reg. 12859 (1971). At that time, the passive belt approved by the agency could not be detached.12 Only later, at a manufacturer’s behest, did the agency approve of the detach-ability feature — and only after assurances that the feature would not compromise the safety benefits of the restraint.13 Although it was then foreseen that 60% of the new cars would contain airbags and 40% would have automatic seatbelts, the ratio between the two was not significant as long as the passive belt would also assure greater passenger safety.
The agency has now determined that the detachable automatic belts will not attain anticipated safety benefits because so many individuals will detach the mechanism. Even if this conclusion were acceptable in its entirety, see infra, at 51-54, standing alone it would not justify any more than an amendment of Standard 208 to disallow compliance by means of the one technology which will not provide effective passenger protection. It does not cast doubt on the need for a passive restraint standard or upon the efficacy of airbag technology. In its most recent rulemaking, the agency again acknowledged the lifesaving potential of the airbag:
[48]*48“The agency has no basis at this time for changing its earlier conclusions in 1976 and 1977 that basic air bag technology is sound and has been sufficiently demonstrated to be effective in those vehicles in current use . . . NHTSA Final Regulatory Impact Analysis (RIA) XI-4 (Oct. 1981), App. 264.
Given the effectiveness ascribed to airbag technology by the agency, the mandate of the Act to achieve traffic safety would suggest that the logical response to the faults of detachable seatbelts would be to require the installation of airbags. At the very least this alternative way of achieving the objectives of the Act should have been addressed and adequate reasons given for its abandonment. But the agency not only did not require compliance through airbags, it also did not even consider the possibility in its 1981 rulemaking. Not one sentence of its rulemaking statement discusses the airbags-only option. Because, as the Court of Appeals stated, “NHTSA’s . . . analysis of airbags was nonexistent,” 220 U. S. App. D. C., at 200, 680 F. 2d, at 236, what we said in Burlington Truck Lines, Inc. v. United States, 371 U. S., at 167, is apropos here:
“There are no findings and no analysis here to justify the choice made, no indication of the basis on which the [agency] exercised its expert discretion. We are not prepared to and the Administrative Procedure Act will not permit us to accept such . . . practice. . . . Expert discretion is the lifeblood of the administrative process, but ‘unless we make the requirements for administrative action strict and demanding, expertise, the strength of modern government, can become a monster which rules with no practical limits on its discretion.’ New York v. United States, 342 U. S. 882, 884 (dissenting opinion)” (footnote omitted).
We have frequently reiterated that an agency must cogently explain why it has exercised its discretion in a given manner, [49]*49Atchison, T. & S. F. R. Co. v. Wichita Bd. of Trade, 412 U. S., at 806; FTC v. Sperry & Hutchinson Co., 405 U. S. 233, 249 (1972); NLRB v. Metropolitan Life Ins. Co., 380 U. S. 438, 443 (1965); and we reaffirm this principle again today.
The automobile industry has opted for the passive belt over the airbag, but surely it is not enough that the regulated industry has eschewed a given safety device. For nearly a decade, the automobile industry waged the regulatory equivalent of war against the airbag14 and lost — the inflatable restraint was proved sufficiently effective. Now the automobile industry has decided to employ a seatbelt system which will not meet the safety objectives of Standard 208. This hardly constitutes cause to revoke the Standard itself. Indeed, the Act was necessary because the industry was not sufficiently responsive to safety concerns. The Act intended that safety standards not depend on current technology and could be “technology-forcing” in the sense of inducing the development of superior safety design. See Chrysler Corp. v. Department of Transportation, 472 F. 2d, at 672-673. If, under the statute, the agency should not defer to the industry’s failure to develop safer cars, which it surely should not do, a fortiori it may not revoke a safety standard which can be satisfied by current technology simply because the industry has opted for an ineffective seatbelt design.
Although the agency did not address the mandatory airbag option and the Court of Appeals noted that “airbags seem to have none of the problems that NHTSA identified in passive seatbelts,” 220 U. S. App. D. C., at 201, 680 F. 2d, at 237, petitioners recite a number of difficulties that they [50]*50believe would be posed by a mandatory airbag standard. These range from questions concerning the installation of airbags in small cars to that of adverse public reaction. But these are not the agency’s reasons for rejecting a mandatory airbag standard. Not having discussed the possibility, the agency submitted no reasons at all. The short — and sufficient — answer to petitioners’ submission is that the courts may not accept appellate counsel’s post hoc rationalizations for agency action. Burlington Truck Lines, Inc. v. United States, 371 U. S., at 168. It is well established that an agency’s action must be upheld, if at all, on the basis articulated by the agency itself. Ibid.; SEC v. Chenery Corp., 332 U. S., at 196; American Textile Mfrs. Institute, Inc. v. Donovan, 452 U. S. 490, 539 (1981).15
Petitioners also invoke our decision in Vermont Yankee Nuclear Power Corp. v. Natural Resources Defense Council, Inc., 435 U. S. 519 (1978), as though it were a talisman under which any agency decision is by definition unimpeachable. Specifically, it is submitted that to require an agency to consider an airbags-only alternative is, in essence, to dictate to the agency the procedures it is to follow. Petitioners both misread Vermont Yankee and misconstrue the nature of the remand that is in order. In Vermont Yankee, we held that a court may not impose additional procedural requirements upon an agency. We do not require today any specific proce[51]*51dures which NHTSA must follow. Nor do we broadly require an agency to consider all policy alternatives in reaching decision. It is true that rulemaking “cannot be found wanting simply because the agency failed to include every alternative device and thought conceivable by the mind of man . . . regardless of how uncommon or unknown that alternative may have been . . . .” Id., at 551. But the airbag is more than a policy alternative to the passive restraint Standard; it is a technological alternative within the ambit of the existing Standard. We hold only that given the judgment made in 1977 that airbags are an effective and cost-beneficial lifesaving technology, the mandatory passive restraint rule may not be abandoned without any consideration whatsoever of an airbags-only requirement.
B
Although the issue is closer, we also find that the agency was too quick to dismiss the safety benefits of automatic seatbelts. NHTSA’s critical finding was that, in light of the industry’s plans to install readily detachable passive belts, it could not reliably predict “even a 5 percentage point increase as the minimum level of expected usage increase.” 46 Fed. Reg. 53423 (1981). The Court of Appeals rejected this finding because there is “not one iota” of evidence that Modified Standard 208 will fail to increase nationwide seatbelt use by at least 13 percentage points, the level of increased usage necessary for the Standard to justify its cost. Given the lack of probative evidence, the court held that “only a well justified refusal to seek more evidence could render rescission non-arbitrary.” 220 U. S. App. D. C., at 196, 680 F. 2d, at 232.
Petitioners object to this conclusion. In their view, “substantial uncertainty” that a regulation will accomplish its intended purpose is sufficient reason, without more, to rescind a regulation. We agree with petitioners that just as an agency reasonably may decline to issue a safety standard if it is uncertain about its efficacy, an agency may also revoke a [52]*52standard on the basis of serious uncertainties if supported by the record and reasonably explained. Rescission of the passive restraint requirement would not be arbitrary and capricious simply because there was no evidence in direct support of the agency’s conclusion. It is not infrequent that the available data do not settle a regulatory issue, and the agency must then exercise its judgment in moving from the facts and probabilities on the record to a policy conclusion. Recognizing that policymaking in a complex society must account for uncertainty, however, does not imply that it is sufficient for an agency to merely recite the terms “substantial uncertainty” as a justification for its actions. As previously noted, the agency must explain the evidence which is available, and must offer a “rational connection between the facts found and the choice made.” Burlington Truck Lines, Inc. v. United States, supra, at 168. Generally, one aspect of that explanation would be a justification for rescinding the regulation before engaging in a search for further evidence.
In these cases, the agency’s explanation for rescission of the passive restraint requirement is not sufficient to enable us to conclude that the rescission was the product of reasoned decisionmaking. To reach this conclusion, we do not upset the agency’s view of the facts, but we do appreciate the limitations of this record in supporting the agency’s decision. We start with the accepted ground that if used, seatbelts unquestionably would save many thousands of lives and would prevent tens of thousands of crippling injuries. Unlike recent regulatory decisions we have reviewed, Industrial Union Dept. v. American Petroleum Institute, 448 U. S. 607 (1980); American Textile Mfrs. Institute, Inc. v. Donovan, 452 U. S. 490 (1981), the safety benefits of wearing seatbelts are not in doubt, and it is not challenged that were those benefits to accrue, the monetary costs of implementing the Standard would be easily justified. We move next to the fact that there is no direct evidence in support of the agency’s finding that detachable automatic belts cannot be predicted [53]*53to yield a substantial increase in usage. The empirical evidence on the record, consisting of surveys of drivers of automobiles equipped with passive belts, reveals more than a doubling of the usage rate experienced with manual belts.16 Much of the agency’s rulemaking statement — and much of the controversy in these cases — centers on the conclusions that should be drawn from these studies. The agency maintained that the doubling of seatbelt usage in these studies could not be extrapolated to an across-the-board mandatory standard because the passive seatbelts were guarded by ignition interlocks and purchasers of the tested cars are somewhat atypical.17 Respondents insist these studies demonstrate that Modified Standard 208 will substantially increase seatbelt usage. We believe that it is within the agency’s discretion to pass upon the generalizability of these field studies. This is precisely the type of issue which rests within the expertise of NHTSA, and upon which a reviewing court must be most hesitant to intrude.
But accepting the agency’s view of the field tests on passive restraints indicates only that there is no reliable real-world experience that usage rates will substantially increase. To be sure, NHTSA opines that “it cannot reliably predict even a 5 percentage point increase as the minimum level of [54]*54expected increased usage.” Notice 25, 46 Fed. Reg. 53423 (1981). But this and other statements that passive belts will not yield substantial increases in seatbelt usage apparently take no account of the critical difference between detachable automatic belts and current manual belts. A detached passive belt does require an affirmative act to reconnect it, but— unlike a manual seatbelt — the passive belt, once reattached, will continue to function automatically unless again disconnected. Thus, inertia — a factor which the agency’s own studies have found significant in explaining the current low usage rates for seatbelts18 — works in favor of, not against, use of the protective device. Since 20% to 50% of motorists currently wear seatbelts on some occasions,19 there would seem to be grounds to believe that seatbelt use by occasional users will be substantially increased by the detachable passive belts. Whether this is in fact the case is a matter for the agency to decide, but it must bring its expertise to bear on the question.
The agency is correct to look at the costs as well as the benefits of Standard 208. The agency’s conclusion that the incremental costs of the requirements were no longer reasonable was predicated on its prediction that the safety benefits of the regulation might be minimal. Specifically, the [55]*55agency’s fears that the public may resent paying more for the automatic belt systems is expressly dependent on the assumption that detachable automatic belts will not produce more than “negligible safety benefits.” Id., at 53424. When the agency reexamines its findings as to the likely increase in seatbelt usage, it must also reconsider its judgment of the reasonableness of the monetary and other costs associated with the Standard. In reaching its judgment, NHTSA should bear in mind that Congress intended safety to be the pre-eminent factor under the Act:
“The Committee intends that safety shall be the overriding consideration in the issuance of standards under this bill. The Committee recognizes . . . that the Secretary will necessarily consider reasonableness of cost, feasibility and adequate leadtime.” S. Rep. No. 1301, 89th Cong., 2d Sess., 6 (1966).
“In establishing standards the Secretary must conform to the requirement that the standard be practicable. This would require consideration of all relevant factors, including technological ability to achieve the goal of a particular standard as well as consideration of economic factors.
“Motor vehicle safety is the paramount purpose of this bill and each standard must be related thereto.” H. R. Rep. No. 1776, 89th Cong., 2d Sess., 16 (1966).
The agency also failed to articulate a basis for not requiring nondetachable belts under Standard 208. It is argued that the concern of the agency with the easy detachability of the currently favored design would be readily solved by a continuous passive belt, which allows the occupant to “spool out” the belt and create the necessary slack for easy extrication from the vehicle. The agency did not separately consider the continuous belt option, but treated it together with the ignition interlock device in a category it titled “Option of Adopting Use-Compelling Features.” 46 Fed. Reg. 53424 [56]*56(1981). The agency was concerned that use-compelling devices would “complicate the extrication of [an] occupant from his or her car.” Ibid. “[T]o require that passive belts contain use-compelling features,” the agency observed, “could be counterproductive [, given]. . . widespread, latent and irrational fear in many members of the public that they could be trapped by the seat belt after a crash.” Ibid. In addition, based on the experience with the ignition interlock, the agency feared that use-compelling features might trigger adverse public reaction.
By failing to analyze the continuous seatbelts option in its own right, the agency has failed to offer the rational connection between facts and judgment required to pass muster under the arbitrary-and-capricious standard. We agree with the Court of Appeals that NHTSA did not suggest that the emergency release mechanisms used in nondetachable belts are any less effective for emergency egress than the buckle release system used in detachable belts. In 1978, when General Motors obtained the agency’s approval to install a continuous passive belt, it assured the agency that nondetachable belts with spool releases were as safe as detachable belts with buckle releases. 43 Fed. Reg. 21912, 21913-21914 (1978). NHTSA was satisfied that this belt design assured easy extricability: “[t]he agency does not believe that the use of [such] release mechanisms will cause serious occupant egress problems . . . .” Id., at 52493, 52494. While the agency is entitled to change its view on the acceptability of continuous passive belts, it is obligated to explain its reasons for doing so.
The agency also failed to offer any explanation why a continuous passive belt would engender the same adverse public reaction as the ignition interlock, and, as the Court of Appeals concluded, “every indication in the record points the other way.” 220 U. S. App. D. C., at 198, 680 F. 2d, at 234.20
[57]*57We see no basis for equating the two devices: the continous belt, unlike the ignition interlock, does not interfere with the operation of the vehicle. More importantly, it is the agency’s responsibility, not this Court’s, to explain its decision.
VI
“An agency’s view of what is in the public interest may change, either with or without a change in circumstances. But an agency changing its course must supply a reasoned analysis . ...” Greater Boston Television Corp. v. FCC, 143 U. S. App. D. C. 383, 394, 444 F. 2d 841, 852 (1970) (footnote omitted), cert. denied, 403 U. S. 923 (1971). We do not accept all of the reasoning of the Court of Appeals but we do conclude that the agency has failed to supply the requisite “reasoned analysis” in this case. Accordingly, we vacate the judgment of the Court of Appeals and remand the cases to that court with directions to remand the matter to the NHTSA for further consideration consistent with this opinion.21
So ordered.