concurring in the judgment.
In Edwards v. Arizona, 451 U. S. 477 (1981), this Court determined that the accused’s waiver of his right to counsel *652during custodial interrogation was involuntary because he was subjected to renewed interrogation without counsel present after having invoked that right. It was uncertain at the time whether the Court merely intended to apply Johnson v. Zerbst, 304 U. S. 458, 464 (1938), that had held that waivers of counsel are effective only if they are “an intentional relinquishment or abandonment of a known right or privilege,” a determination made by reference to “the particular facts and circumstances surrounding [each] case, including the background, experience, and conduct of the accused.” See 451 U. S., at 482. Alternatively, Edwards could have been interpreted as establishing a new per se rule that, once the right to counsel has been invoked, a waiver of that right, however voluntary under the Zerbst standard, can never be valid if made in response to further police questioning. See Edwards, supra, at 488-490 (Powell, J., concurring in result). Confusion as to the proper interpretation of Edwards persisted in subsequent cases. See, e. g., Oregon v. Bradshaw, 462 U. S. 1039 (1983); id., at 1047, n. 1 (Powell, J., concurring in judgment) (citing lower court cases). The Court now states clearly, relying in part on Bradshaw, that Edwards established a new per se rule and to that extent overruled Johnson v. Zerbst, supra.1 Ante, at 647-648.
This acknowledgment suffices, in my view, to resolve the issue posed by the present case. I previously have urged the Court to adopt Justice Harlan’s suggestion that a new rule of constitutional law should be applied only to review *653of criminal convictions not yet final when the rule is announced.2 Hankerson v. North Carolina, 432 U. S. 233, 246-248 (1977) (concurring in judgment). As Justice Harlan reasoned in Mackey v. United States, 401 U. S. 667, 675-695 (1971) (concurring in judgments in part and dissenting in part), that approach follows directly from a proper conception of the scope of the writ of habeas corpus, as contrasted to direct review. A brief review of the reasons for that approach relevant to the present case will explain why I do not join the Court’s opinion.
Retroactive application on habeas corpus of constitutional rules governing criminal procedure is unnecessary to advance the purposes of habeas corpus, even under a regime that permits the federal courts on habeas to vacate a final conviction on any properly preserved ground of federal constitutional error. Review on habeas to determine that the conviction rests upon correct application of the law in effect at the time of the conviction is all that is required to “forc[e] trial and appellate courts ... to toe the constitutional mark.”3 Id., at 687. Nor will fundamental fairness require complete retroactivity, except in rare instances.4 Because retroactive *654application of new rules of constitutional law generally does little to advance the purposes of collateral relief on habeas, it is particularly difficult in such cases to justify imposing upon the State the costs of collateral review. These are not insubstantial. They include “the burden on judicial and prosecu-torial resources entailed in retrial” and “the miscarriage of justice that occurs when a guilty offender is set free only because effective retrial is impossible years after the offense.” Hankerson v. North Carolina, supra, at 247. Retroactive application of constitutional rules frustrates the State’s enforcement of its criminal law despite the State’s careful adherence to the federal constitutional standards that governed at the time of the prisoner’s conviction.
The costs imposed upon the State by retroactive application of new rules of constitutional law on habeas corpus thus generally far outweigh the benefits of this application. It is therefore unnecessary to consider the Linkletter/Stovall factors, as these were intended merely to guide the Court’s balancing of the costs and benefits that accrue from retroactive application of a particular rule.
Certainly the per se test adopted in Edwards is not a rule necessary to assure fundamental fairness. As the Court’s opinion states, “in those situations where renewed interrogation raises significant doubt as to the voluntariness and reliability of the statement and, therefore, the accuracy of the outcome at trial, it is likely that suppression could be achieved without reliance on the prophylactic rule adopted in Edwards” Ante, at 644.
For these reasons, I concur in the judgment.
In Edwards, although concurring in the judgment, I expressed concern as to whether there was an intent to overrule Zerbst. See 451 U. S., at 491-492. In Bradshaw, last Term, in an opinion also concurring only in the judgment, I reiterated my conviction that the Constitution requires no per se rule on an issue as purely factual as whether and when a valid waiver of counsel occurs. 462 U. S., at 1049-1051. As the contrasting opinions of Justices Marshall and Rehnquist in Bradshaw illustrate, even the new per se rule is more likely to confuse than to clarify. See id., at 1048 (Powell, J., concurring in judgment). Nevertheless, I now, of course, accept Edwards and Bradshaw as binding authority.
The Court adopted this view in United States v. Johnson, 457 U. S. 537 (1982), to the extent of holding that new rules of Fourth Amendment law would be applied to all convictions not yet final when the rule was announced.
Although it might seem desirable perpetually to revise past convictions in light of evolving legal doctrine, the attempt to do so is fundamentally at odds with the rule of law. “At some point, the criminal process, if it is to function at all, must turn its attention from whether a man ought properly to be incarcerated to how he is to be treated once convicted. If law, criminal or otherwise, is worth having and enforcing, it must at some time provide a definitive answer to the questions litigants present or else it never provides an answer at all.” Mackey v. United States, 401 U. S., at 690-691 (opinion of Harlan, J.). See also Schneckloth v. Bustamonte, 412 U. S. 218, 262 (1973) (POWELL, J., concurring).
We should give retroactive effect on habeas to decisions announcing rules of criminal procedure required to ensure fundamental fairness, e. g., Gideon v. Wainwright, 372 U. S. 335 (1963), or holding conduct entirely *654immune from criminal punishment, e. g., Roe v. Wade, 410 U. S. 113 (1973). Releasing on habeas prisoners who have been convicted by fundamentally unfair procedures, or who have committed no constitutionally punishable offense at all, would give effect to our decisions in those rare eases where a conviction fully in accord with the law governing at the time of conviction is nonetheless plainly unjust. See Mackey v. United States, supra, at 692-693.