with whom The Chief Justice and Justice Rehnquist join, dissenting.
The Court today holds Hamilton County liable for the forcible entry in May 1977 by Deputy Sheriffs into petitioner’s office. The entry and subsequent search were pursuant to capiases for third parties — petitioner’s employees — who had failed to answer a summons to appear as witnesses before a grand jury investigating petitioner. When petitioner refused to allow the Sheriffs to enter, one of them, at the request of his supervisor, called the office of the County Prosecutor for instructions. The Assistant County Prosecutor received the call, and apparently was in doubt as to what advice to give. He referred the question to the County Prosecutor, who advised the Deputy Sheriffs to “go in and get them [the witnesses]” pursuant to the capiases. This five-word response to a single question over the phone is now found by this Court to have created an official county policy for which Hamilton County is liable under § 1983. This holding is wrong for at least two reasons. First, the Prosecutor’s response and the Deputies’ subsequent actions did not violate any constitutional right that existed at the time of the forcible entry. Second, no official county policy could have been created solely by an offhand telephone response from a busy County Prosecutor.
» — I
Petitioner s allegation of a constitutional violation rests exclusively on Steagald v. United States, 451 U. S. 204 (1981), decided four years after the entry here. In Steagald we held that an officer may not search for the subject of an arrest warrant in a third party’s home without first obtaining a search warrant, unless the search is consensual or justified by exigent circumstances. In 1977, the law in the Sixth Circuit was that a search warrant was not required in such situations if the police had an arrest warrant and reason to believe that the person to be arrested was within the home to be *493searched. United States v. McKinney, 379 F. 2d 259, 262-263 (1967). That view was shared by at least two other Circuits. See United States v. Gaultney, 606 F. 2d 540, 544-545 (CA5 1979); United States v. Harper, 550 F. 2d 610, 612-614 (CA10), cert. denied, 434 U. S. 837 (1977). Another Circuit had favored that view in dicta. See United States v. Manley, 632 F. 2d 978, 983 (CA2 1980). Thus, under the governing law in the applicable Circuit, uncontradicted by any opinion of this Court, the entry into petitioner’s office pursuant to an arrest warrant was not a violation of petitioner’s Fourth Amendment rights. .
The only way to transform this search — legitimate at the time — into a constitutional violation is to apply Steagald retroactively. This would not be a startling proposition if all that petitioner sought was retroactive application of a new rule of criminal law to a direct appeal from his criminal conviction.1 But petitioner seeks something very different— retroactive application of the new rule of criminal law announced in Steagald to his subsequent civil lawsuit. Even if one accepts the proposition that a new rule of criminal law should be applied retroactively to create a basis for civil liability under § 1983,2 existing principles of retroactivity for *494civil cases show that Steagald should not be applied retroactively to this action.
The leading case explaining the framework of analysis for civil retroactivity is Chevron Oil Co. v. Huson, 404 U. S. 97 (1971). Under Chevron, a court must assess: (i) whether the new decision “established] a new principle of law ... by overruling clear past precedent... or by deciding an issue of first impression whose resolution was not clearly foreshadowed,” id., at 106; (ii) “the prior history of the rule in question, its purpose and effect, and whether retrospective operation will further or retard its operation,” id., at 107; and (iii) the respective inequities of retroactive or nonretroactive application, ibid.
When viewed in light of these factors, retroactive application of Steagald is not justified. First, Steagald overruled past Courts of Appeals precedent, and the decision had not been foreshadowed in opinions of this Court. The governing law in three Federal Circuits permitted searches of third parties’ homes pursuant to an arrest warrant, see swpra, at 493, and earlier decisions of this Court arguably supported such searches.3 Second, the “purpose” of Steagald was to clarify the application of the Fourth Amendment to such searches, not to provide for money damages. Finally, retro*495active application of Steagald in this context would produce substantial inequitable results by imposing liability on local government units for law enforcement practices that were legitimate at the time they were undertaken. See Griffin v. Illinois, 351 U. S. 12, 26 (1956) (Frankfurter, J., concurring in judgment) (“We should not indulge in the fiction that the law now announced has always been the law . . .”)• Civil liability should not attach unless there was notice that a constitutional right was at risk. Procunier v. Navarette, 434 U. S. 555, 562 (1978).
We ought to be even more wary of applying new rules of Fourth Amendment law retroactively to civil cases than we are with new rules of civil law. The primary reason for imposing §1983 liability on local government units is deterrence, so that if there is any doubt about the constitutionality of their actions, officials will “err on the side of protecting citizens’ rights.” Owen v. City of Independence, 445 U. S. 622, 652 (1980). But law enforcement officials, particularly prosecutors, are in a much different position with respect to deterrence than other local government officials. Cf. Imbler v. Pachtman, 424 U. S. 409, 425 (1976). Their affirmative duty to enforce the law vigorously often requires them to take actions that legitimately intrude on individual liberties, often acting “under serious constraints of time and even information.” Ibid. While law enforcement officials, as much as any other official, ought to “err on the side of protecting citizens’ rights” when they have legitimate doubts about the constitutionality of their actions, they should not be deterred from doing their duty to enforce the criminal law when they have no such doubts. In this case, for example, Sixth Circuit law expressly authorized the Prosecutor’s decision. Because a court engages in the same balancing of interests in a Fourth Amendment case that is required, with much less deliberation, of law enforcement officials, they are justified in relying on the judgment of the applicable federal court. Under these circumstances, there was nothing that should *496have caused the officials to “harbor doubts about the lawfulness of their intended actions,” Owen, supra, at 652, and therefore nothing to deter.
Moreover, there is a significant cost to unwarranted deterrence of law enforcement officials. We recognized in Irnbler a strong state interest in “vigorous and fearless” prosecution, and found that to be “essential to the proper functioning of the criminal justice system.” 424 U. S., at 427-428. Those same general concerns apply to other law enforcement officials. Unwarranted deterrence has the undesirable effect of discouraging conduct that is essential to our justice system and protects the State’s interest in public safety. In that sense, this case is different from Owen. It is no answer to say that some officials are entitled either to absolute or qualified immunity. It ignores reality to say that if petitioner is successful in his $20-million suit it will not have a chilling effect on law enforcement practices in Hamilton County.4
For these reasons, Steagald should not be applied retroactively. Consequently, petitioner has no constitutional violation of which to complain. I therefore would affirm the decision of the Court of Appeals.5
*497II
Even if Steagald is applied retroactively, petitioner has failed to demonstrate the existence of an official policy for which Hamilton County can be liable. The action said to have created policy here was nothing more than a brief response to a single question over the telephone. The Deputy Sheriffs sought instructions concerning a situation that had never occurred before, at least in the memory of the participants. Ante, at 474-475. That in itself, and the fact that the Assistant Prosecutor had to obtain advice from the County Prosecutor, strongly indicate that no prior policy had been formed. Petitioner therefore argues that the County Prosecutor’s reaction in this case formed county policy. The sparse facts supporting petitioner’s theory — adopted by the Court today — do not satisfy the requirement in Monell v. New York City Dept. of Social Services, 436 U. S. 658, 691 (1978), that local government liability under §1983 be imposed only when the injury is caused by government policy.
A
Under Monell, local government units may be liable only when “the action that is alleged to be unconstitutional implements or executes a policy statement, ordinance, regulation, or decision officially adopted and promulgated by that body’s officers.” 436 U. S., at 690. This case presents the opportunity to define further what was meant in Monell by “official policy.” Proper resolution of the case calls for identification of the applicable principles for determining when policy is *498created. The Court today does not do this, but instead focuses almost exclusively on the status of the decisionmaker. Its reasoning is circular: it contends that policy is what policymakers make, and policymakers are those who have authority to make policy.
The Court variously notes that if a decision “is properly made by that government’s authorized decisionmakers, it surely represents an act of official government ‘policy’ as that term is commonly understood,” ante, at 481, and that “where action is directed by those who establish governmental policy, the municipality is equally responsible . . . ,” ibid. Thus, the Court’s test for determining the existence of policy focuses only on whether a decision was made “by the official or officials responsible for establishing final policy with respect to the subject matter in question.” Ante, at 483-484.
In my view, the question whether official policy — in any normal sense of the term — has been made in a particular case is not answered by explaining who has final authority to make policy. The question here is not “could the County Prosecutor make policy?” but rather, “did he make policy?” By focusing on the authority granted to the official under state law, the Court’s test fails to answer the key federal question presented. The Court instead turns the question into one of state law. Under a test that focuses on the authority of the decisionmaker, the Court has only to look to state law for the resolution of this case. Here the Court of Appeals found that “both the County Sheriff and the County Prosecutor [had authority under Ohio law to] establish county policy under appropriate circumstances.” Ante, at 484. Apparently that recitation of authority is all that is needed under the Court’s test because no discussion is offered to demonstrate that the Sheriff or the Prosecutor actually used that authority to establish official county policy in this case.
*499Moreover, the Court’s reasoning is inconsistent with Monell. Today’s decision finds that policy is established because a policymaking official made a decision on the telephone that was within the scope of his authority. The Court ignores the fact that no business organization or governmental unit makes binding policy decisions so cavalierly. The Court provides no mechanism for distinguishing those acts or decisions that cannot fairly be construed to create official policy from the normal process of establishing an official policy that would be followed by a responsible public entity. Thus, the Court has adopted in part what it rejected in Monell: local government units are now subject to respondeat superior liability, at least with respect to a certain category of employees, i. e., those with final authority to make policy. See Monell, 436 U. S., at 691; see also Oklahoma City v. Tuttle, 471 U. S. 808, 818 (1985) (rejecting theories akin to respondeat superior) (plurality opinion). The Court’s reliance on the status of the employee carries the concept of “policy” far beyond what was envisioned in Monell.
B
In my view, proper resolution of the question whether official policy has been formed should focus on two factors: (i) the nature of the decision reached or the action taken, and (ii) the process by which the decision was reached or the action was taken.
Focusing on the nature of the decision distinguishes between policies and mere ad hoc decisions. Such a focus also reflects the fact that most policies embody a rule of general applicability. That is the tenor of the Court’s statement in Monell that local government units are liable under § 1983 when the action that is alleged to be unconstitutional “implements or executes a policy statement, ordinance, regulation, or decision officially adopted and promulgated by that body’s officers.” 436 U. S., at 690. The clear implication is that policy is created when a rule is formed that applies to all simi*500lar situations — a “governing principle [or] plan.” Webster’s New Twentieth Century Dictionary 1392 (2d ed. 1979).6 When a rule of general applicability has been approved, the government has taken a position for which it can be held responsible.7
Another factor indicating that policy has been formed is the process by which the decision at issue was reached. Formal procedures that involve, for example, voting by elected officials, prepared reports, extended deliberation, or official records indicate that the resulting decisions taken “may fairly be said to represent official policy,” Monell, supra, at 694. Owen v. City of Independence, 445 U. S. 622 (1980), provides an example. The City Council met in a regularly scheduled meeting. One member of the Council made a motion to release to the press certain reports that cast an employee in a bad light. After deliberation, the Council passed the motion with no dissents and one abstention. Id., at 627-629. Although this official action did not establish a rule of general applicability, it is clear that policy was formed because of the process by which the decision was reached.
Applying these factors to the instant case demonstrates that no official policy was formulated. Certainly, no rule of general applicability was adopted. The Court correctly *501notes that the Sheriff “testified that the Department had no written policy respecting the serving of capiases on the property of third persons and that the proper response in any given situation would depend upon the circumstances.” Ante, at 474. Nor could he recall a specific instance in which entrance had been denied and forcibly gained. The Court’s result today rests on the implicit conclusion that the Prosecutor’s response — “go in and get them” — altered the prior case-by-case approach of the Department and formed a new rule to apply in all similar cases. Nothing about the Prosecutor’s response to the inquiry over the phone, nor the circumstances surrounding the response, indicates that such a rule of general applicability was formed.8
Similarly, nothing about the way the decision was reached indicates that official policy was formed. The prosecutor, without time for thoughtful consideration or consultation, simply gave an off-the-cuff answer to a single question. There was no process at all. The Court’s holding undercuts *502the basic rationale of Monell, and unfairly increases the risk of liability on the level of government least able to bear it. I dissent.
In fact, on direct appeal from his criminal conviction, petitioner did enjoy retroactive application of the rule in Steagald, although it did not entitle him to reversal of his conviction. State v. Pembaur, 9 Ohio St. 3d 136, 459 N. E. 2d 217 (1984). While the Ohio Supreme Court did not specifically address the retroactivity issue, it did discuss the applicability of Steagald to petitioner’s criminal appeal. 9 Ohio St. 3d, at 137-138, 459 N. E. 2d, at 218-219. The court reasoned, however, that because no “substantive” offense was involved, but only a conviction for obstructing the police, petitioner could not rely on the uneonstitutionality of the search as a defense. Id., at 138, 459 N. E. 2d, at 219.
If new criminal rules are so applied, it is possible that a person could obtain the benefit of retroactive application of a new criminal rule to his civil § 1983 case, even though he could not use the new rule to attack his conviction collaterally. A prisoner literally could be forced to remain in prison while collecting his civil damages award. In Shea v. Louisiana, 470 U. S. 51 (1985), the Court created a distinction between retroactivity on *494direct review of a conviction and on collateral attack of a conviction that has become final. On collateral attack the principles of Solern v. Stumes, 465 U. S. 638 (1984), apply, which make it less likely that a new rule would be applied retroactively. A key factor under Stumes is the extent of the reliance by law enforcement authorities on the old standards. Id., at 643.
In Dalia v. United States, 441 U. S. 238 (1979), the Court rejected the argument that a separate search warrant was required before police could enter a business office to install an eavesdropping device when officers already had a warrant authorizing the eavesdropping itself. The Court noted that “in executing a warrant the police may find it necessary to interfere with privacy rights not explicitly considered by the judge who issued the warrant.” Id., at 257. In Payton v. New York, 445 U. S. 573 (1980), the Court rejected the suggestion that a separate search warrant was required before police could execute an arrest warrant by entering the home of the subject of the warrant.
Justice Stevens misunderstands the unique posture of this case. This is not a question of retroactivity of a new civil rule to civil cases versus retroactivity of a new criminal rule to criminal cases. The special concerns discussed in the text above arise in part out of the retroactive application of a new rule of criminal law to civil eases. I see little to be gained by comparing the societal costs of civil and criminal retroactivity, see concurring opinion of Stevens, J., ante, at 488-489, n. 3, because they can be severe in either case. Today’s opinion could result in even a nonnegligent mistake in judgment imposing heavy liability on units of local government, especially now in view of the skyrocketing cost — or unavailability — of liability insurance. See also Malley v. Briggs, ante, p. 335.
The Court’s only response to these concerns is to note that respondent has “never challenged and has in fact also conceded that Steagald applied retroactively to this case. ... We decide this case in light of respondent’s concessions.” Ante, at 477, n. 5. The retroactivity issue, however, is central to this case. We need not reach the difficult federal issues in this case if the Court correctly resolved Steagald’s retroactivity. *497Nor are we prevented from doing so by any actual concession of the respondent. See Tr. of Oral Arg. 26-27. Justice White does not address the retroactivity of Steagald on the ground that the county had not relied on this contention. In my view, although we need not address this retro-activity issue, there is no question as to our right to do so — especially in view of the unfairness of holding the respondent county liable for not anticipating Steagald. Procunier v. Navarette, 434 U. S. 555, 559, n. 6 (1978); Blonder-Tongue Laboratories, Inc. v. University of Illinois Foundation, 402 U. S. 313, 320, n. 6 (1971).
The focus on a rule of general applicability does not mean that more than one instance of its application is required. The local government unit may be liable for the first application of a duly constituted unconstitutional policy.
An example of official policy in the form of a rule of general applicability is Newport v. Fact Concerts, Inc., 453 U. S. 247 (1981). While the Court in that case was not called on to define the scope of the word “policy,” the complaint was based on a rule of general applicability. The city canceled a scheduled concert pursuant to its rule of not allowing rock concerts. Plaintiffs alleged that the general rule against rock concerts violated their First Amendment rights. Even if the cancellation was the first implementation of the rule, it was clear that the city had committed itself to a general position that would govern future cases.
There is nothing in the record to support the inference relied on by Justices White and O’Connor. Nor has this Court ever held that because a policy has been adopted by one city or county we may assume that a similar policy has been adopted by neighboring cities or counties. After all, the city and county in this case are separate governmental entities.
Moreover, and again contrary to the views of my colleagues, this Court has never held — at least to my knowledge — that we may assume that simply because certain conduct is permitted by existing law, it must have been adopted as county policy. The undisputed facts in this case refute these assumptions by Justices White and O’CONNOR. Neither the Sheriffs who had been denied entry nor the Assistant County Prosecutor knew of any such policy. Otherwise, one of the Sheriffs would not have called the Prosecutor’s office, and certainly the Assistant Prosecutor would not have thought it necessary to put the question to the Prosecutor. Nor did the Prosecutor, when asked, say that the county’s policy was to force an entry when necessary to serve a valid arrest warrant. Rather, he simply said “go in and get them” — the sort of spontaneous reply that a busy official might make quite thoughtlessly. As noted above, the Sheriff testified that the proper response would depend on the circumstances.