delivered the opinion of the Court.

This ease calls for determinations parallel in some respects to those discussed in today’s opinion in Harrington v. Richter, ante, p. 86. Here, as in Richter, the Court reviews a decision of the Court of Appeals for the Ninth Circuit granting federal habeas corpus relief in a challenge to a state criminal conviction. Here, too, the case turns on the proper implementation of one of the stated premises for issuance of federal habeas corpus contained in 28 U. S. C. § 2254(d), the instruction that federal habeas corpus relief may not be granted with respect to any claim a state court has adjudicated on the merits unless, among other exceptions, the state court’s decision denying relief involves “an unreasonable application” of “clearly established Federal law, as determined by the Supreme Court of the United States.” And, as in Richter, the relevant clearly established law derives from Strickland v. Washington, 466 U. S. 668 (1984), which provides the standard for inadequate assistance of counsel under the Sixth Amendment. Richter involves a California conviction and addresses the adequacy of representation when counsel did not consult or use certain experts in pretrial preparation and at trial. The instant case involves an unrelated Oregon conviction and concerns the adequacy of representation in providing an assessment of a plea bargain without first seeking suppression of a confession assumed to have been improperly obtained.

I

On December 7, 1995, respondent Randy Moore and two confederates attacked Kenneth Rogers at his home and bloodied him before tying him with duct tape and throwing[*119] him in the trunk of a car. They drove into the Oregon countryside, where Moore shot Rogers in the temple, killing him.

Afterwards, Moore and one of his accomplices told two people — Moore’s brother and the accomplice’s girlfriend— about the crimes. According to Moore’s brother, Moore and his accomplice admitted:

“[T]o make an example and put some scare into Mr. Rogers . . . , they had blind-folded him [and] duct taped him and put him in the trunk of the car and took him out to a place that’s a little remote .... [T]heir intent was to leave him there and make him walk home ....

“[Moore] had taken the revolver from Lonnie and at the time he had taken it, Mr. Rogers had slipped backwards on the mud and the gun discharged.” App. 157-158.

Moore and his accomplice repeated this account to the police. On the advice of counsel Moore agreed to plead no contest to felony murder in exchange for a sentence of 300 months, the minimum sentence allowed by law for the offense.

Moore later filed for postconviction relief in an Oregon state court, alleging that he had been denied his right to effective assistance of counsel. He complained that his lawyer had not filed a motion to suppress his confession to police in advance of the lawyer’s advice that Moore considered before accepting the plea offer. After an evidentiary hearing, the Oregon court concluded a “motion to suppress would have been fruitless” in light of the other admissible confession by Moore, to which two witnesses could testify. Id., at 140. As the court noted, Moore’s trial counsel explained why he did not move to exclude Moore’s confession to police:

“Mr. Moore and I discussed the possibility of filing a Motion to Suppress and concluded that it would be unavailing, because... he had previously made a full confession to his brother and to [his accomplice’s girlfriend], either[*120] one of whom could have been called as a witness at any time to repeat his confession in full detail.” Jordan Affidavit (Feb. 26,1999), App. to Pet. for Cert. 70, ¶4.

Counsel added that he had made Moore aware of the possibility of being charged with aggravated murder, which carried a potential death sentence, as well as the possibility of a sentence of life imprisonment without parole. See Ore. Rev. Stat. § 163.105(l)(a) (1995). The intense and serious abuse to the victim before the shooting might well have led the State to insist on a strong response. In light of these facts the Oregon court concluded Moore had not established ineffective assistance of counsel under Strickland.

Moore filed a petition for habeas corpus in the United States District Court for the District of Oregon, renewing his ineffective-assistance claim. The District Court denied the petition, finding sufficient evidence to support the Oregon court’s conclusion that suppression would not have made a difference.

A divided panel of the United States Court of Appeals for the Ninth Circuit reversed. Moore v. Czerniak, 574 F. 3d 1092 (2009). In its view the state court’s conclusion that counsel’s action did not constitute ineffective assistance was an unreasonable application of clearly established law in light of Strickland and was contrary to Arizona v. Fulminante, 499 U. S. 279 (1991). Six judges dissented from denial of rehearing en banc. 574 F. 3d, at 1162.

We granted certiorari sub nom. Belleque v. Moore, 559 U. S. 1004 (2010).

II

The statutory authority of federal courts to issue habeas corpus relief for persons in state custody is defined by 28 U. S. C. § 2254, as amended by the Antiterrorism and Effective Death Penalty Act of 1996 (AEDPA). The text of § 2254(d) states:

“An application for a writ of habeas corpus on behalf of a person in custody pursuant to the judgment of a[*121] State court shall not be granted with respect to any claim that was adjudicated on the merits in State court proceedings unless the adjudication of the claim—

“(1) resulted in a decision that was contrary to, or involved an unreasonable application of, clearly established Federal law, as determined by the Supreme Court of the United States; or

“(2) resulted in a decision that was based on an unreasonable determination of the facts in light of the evidence presented in the State court proceeding.”

AEDPA prohibits federal habeas relief for any claim adjudicated on the merits in state court, unless one of the exceptions listed in § 2254(d) obtains. Relevant here is § 2254(d)(l)’s exception “permitting relitigation where the earlier state decision resulted from an 'unreasonable application of’ clearly established federal law.” Richter, ante, at 100. The applicable federal law consists of the rules for determining when a criminal defendant has received inadequate representation as defined in Strickland.

To establish ineffective assistance of counsel “a defendant must show both deficient performance by counsel and prejudice.” Knowles v. Mirzayance, 556 U. S. 111, 122 (2009). In addressing this standard and its relationship to AEDPA, the Court today in Richter, ante, at 104-105, gives the following explanation:

“To establish deficient performance, a person challenging a conviction must show that ‘counsel’s representation fell below an objective standard of reasonableness.’ [Strickland,] 466 U. S., at 688. A court considering a claim of ineffective assistance must apply a 'strong presumption’ that counsel's representation was within the ‘wide range’ of reasonable professional assistance. Id., at 689. The challenger’s burden is to show ‘that counsel made errors so serious that counsel was[*122] not functioning as the “counsel” guaranteed the defendant by the Sixth Amendment.’ Id., at 687.

“With respect to prejudice, a challenger must demonstrate ‘a reasonable probability that, but for counsel’s unprofessional errors, the result of the proceeding would have been different.’. . .

“ ‘Surmounting Strickland’s high bar is never an easy task.’ Padilla v. Kentucky, 559 U. S. 356, 371 (2010). An ineffective-assistance claim can function as a way to escape rules of waiver and forfeiture and raise issues not presented at trial [or in pretrial proceedings], and so the Strickland standard must be applied with scrupulous care, lest ‘intrusive post-trial inquiry’ threaten the integrity of the very adversary process the right to counsel is meant to serve. Strickland, 466 U. S., at 689-690. Even under de novo review, the standard for judging counsel’s representation is a most deferential one. Unlike a later reviewing court, the attorney observed the relevant proceedings, knew of materials outside the record, and interacted with the client, with opposing counsel, and with the judge. It is ‘all too tempting’ to ‘second-guess counsel's assistance after conviction or adverse sentence.’ Id., at 689; see also Bell v. Cone, 535 U. S. 685, 702 (2002); Lockhart v. Fretwell, 506 U. S. 364, 372 (1993). The question is whether an attorney’s representation amounted to incompetence under ‘prevailing professional norms,’ not whether it deviated from best practices or most common custom. Strickland, 466 U. S., at 690.

“Establishing that a state court’s application of Strickland was unreasonable under § 2254(d) is all the more difficult. The standards created by Strickland and § 2254(d) are both ‘highly deferential,’ id., at 689; Lindh v. Murphy, 521 U. S. 320, 333, n. 7 (1997), and when the two apply in tandem, review is ‘doubly’ so, Knowles, 556 U. S., at 123. The Strickland standard is[*123] a general one, so the range of reasonable applications is substantial. 556 U. S., at 123. Federal habeas courts must guard against the danger of equating unreasonableness under Strickland with unreasonableness under § 2254(d). When § 2254(d) applies, the question is not whether counsers actions were reasonable. The question is whether there is any reasonable argument that counsel satisfied Strickland’s deferential standard.”

Ill

The question becomes whether Moore’s counsel provided ineffective assistance by failing to seek suppression of Moore’s confession to police before advising Moore regarding the plea. Finding that any “motion to suppress would have been fruitless,” the state postconviction court concluded that Moore had not received ineffective assistance of counsel. App. 140. The state court did not specify whether this was because there was no deficient performance under Strickland or because Moore suffered no Strickland prejudice, or both. To overcome the limitation imposed by § 2254(d), the Court of Appeals had to conclude that both findings would have involved an unreasonable application of clearly established federal law. See Richter, ante, at 109-110. In finding that this standard was met, the Court of Appeals erred, for the state-court decision was not an unreasonable application of either part of the Strickland rule.

A

The Court of Appeals was wrong to accord scant deference to counsel’s judgment, and doubly wrong to conclude it would have been unreasonable to find that the defense attorney qualified as counsel for Sixth Amendment purposes. Knowles, supra, at 123; Strickland, supra, at 687. Counsel gave this explanation for his decision to discuss the plea bargain without first challenging Moore’s confession to the police: that suppression would serve little purpose in[*124] light of Moore’s other full and admissible confession, to which both his brother and his accomplice’s girlfriend could testify. The state court would not have been unreasonable to accept this explanation.

Counsel also justified his decision by asserting that any motion to suppress was likely to fail. Reviewing the reasonableness of that justification is complicated by the possibility that petitioner forfeited one argument that would have supported its position: The Court of Appeals assumed that a motion would have succeeded because the warden did not argue otherwise. Of course that is not the same as a concession that no competent attorney would think a motion to suppress would have failed, which is the relevant question under Strickland. See Kimmelman v. Morrison, 477 U. S. 365, 382 (1986); Richter, ante, at 109-110. It is unnecessary to consider whether counsel’s second justification was reasonable, however, since the first and independent explanation— that suppression would have been futile — confirms that his representation was adequate under Strickland, or at least that it would have been reasonable for the state court to reach that conclusion.

Acknowledging guilt and accepting responsibility by an early plea respond to certain basic premises in the law and its fimetion. Those principles are eroded if a guilty plea is too easily set aside based on facts and circumstances not apparent to a competent attorney when actions and advice leading to the plea took place. Plea bargains are the result of complex negotiations suffused with uncertainty, and defense attorneys must make careful strategic choices in balancing opportunities and risks. The opportunities, of course, include pleading to a lesser charge and obtaining a lesser sentence, as compared with what might be the outcome not only at trial but also from a later plea offer if the case grows stronger and prosecutors find stiffened resolve. A risk, in addition to the obvious one of losing the chance for a defense verdict, is that an early plea bargain might come[*125] before the prosecution finds its case is getting weaker, not stronger. The State’s case can begin to fall apart as stories change, witnesses become unavailable, and new suspects are identified.

These considerations make strict adherence to the Strickland standard all the more essential when reviewing the choices an attorney made at the plea bargain stage. Failure to respect the latitude Strickland requires can create at least two problems in the plea context. First, the potential for the distortions and imbalance that can inhere in a hindsight perspective may become all too real. The art of negotiation is at least as nuaneed as the art of trial advocacy, and it presents questions further removed from immediate judicial supervision. There are, moreover, special difficulties in evaluating the basis for counsel’s judgment: An attorney often has insights borne of past dealings with the same prosecutor or court, and the record at the pretrial stage is never as full as it is after a trial. In determining how searching and exacting their review must be, habeas courts must respect their limited role in determining whether there was manifest deficiency in light of information then available to counsel. Lockhart v. Fretwell, 506 U. S. 364, 372 (1993). AEDPA compounds the imperative of judicial caution.

Second, ineffective-assistance claims that lack necessary foundation may bring instability to the very process the inquiry seeks to protect. Strickland allows a defendant “to escape rules of waiver and forfeiture,” Richter, ante, at 105. Prosecutors must have assurance that a plea will not be undone years later because of infidelity to the requirements of AEDPA and the teachings of Strickland. The prospect that a plea deal will afterwards be unraveled when a court second-guesses counsel’s decisions while failing to accord the latitude Strickland mandates or disregarding the structure dictated by AEDPA could lead prosecutors to forgo plea bargains that would benefit defendants, a result favorable to no one.

[*126] Whether before, during, or after trial, when the Sixth Amendment applies, the formulation of the standard is the same: reasonable competence in representing the accused. Strickland, 466 U. S., at 688. In applying and defining this standard substantial deference must be accorded to counsel’s judgment. Id., at 689. But at different stages of the case that deference may be measured in different ways.

In the case of an early plea, neither the prosecution nor the defense may know with much certainty what course the case may take. It follows that each side, of necessity, risks consequences that may arise from contingencies or circumstances yet unperceived. The absence of a developed or an extensive record and the circumstance that neither the prosecution nor the defense case has been well defined create a particular risk that an after-the-fact assessment will run counter to the deference that must be accorded counsel’s judgment and perspective when the plea was negotiated, offered, and entered.

Prosecutors in the present case faced the cost of litigation and the risk of trying their case without Moore’s confession to the police. Moore’s counsel could reasonably believe that a swift plea bargain would allow Moore to take advantage of the State’s aversion to these hazards. And whenever cases involve multiple defendants, there is a chance that prosecutors might convince one defendant to testify against another in exchange for a better deal. Moore’s plea eliminated that possibility and ended an ongoing investigation. Delaying the plea for further proceedings would have given the State time to uncover additional incriminating evidence that could have formed the basis of a capital prosecution. It must be remembered, after all, that Moore’s claim that it was an accident when he shot the victim through the temple might be disbelieved.

It is not clear how the successful exclusion of the confession would have affected counsel’s strategic calculus. The prosecution had at its disposal two witnesses able to relate[*127] another confession. True, Moore’s brother and the girlfriend of his accomplice might have changed their accounts in a manner favorable to Moore. But the record before the state cpurt reveals no reason to believe that either witness would violate the legal obligation to convey the content of Moore’s confession. And to the extent that his accomplice’s girlfriend had an ongoing interest in the matter, she might have been tempted to put more blame, not less, on Moore. Then, too, the accomplices themselves might have decided to implicate Moore to a greater extent than his own confession did, say by indicating that Moore shot the victim deliberately, not accidentally. All these possibilities are speculative. What counsel knew at the time was that the existence of the two witnesses to an additional confession posed a serious strategic concern.

Moore’s prospects at trial were thus anything but certain. Even now, he does not deny any involvement in the kidnaping and killing. In these circumstances, and with a potential capital charge lurking, Moore’s counsel made a reasonable choice to opt for a quick plea bargain. At the very least, the state court would not have been unreasonable to so conclude. Cf. Yarborough v. Alvarado, 541 U. S. 652, 664 (2004) (explaining that state courts enjoy “more leeway” under AEDPA in applying general standards).

The Court of Appeals’ contrary holding rests on a case that did not involve ineffective assistance of counsel: Arizona v. Fulminante, 499 U. S. 279. To reach that result, it transposed that case into a novel context; and novelty alone — at least insofar as it renders the relevant rule less than “clearly established” — provides a reason to reject it under AEDPA. See Yarborough, supra, at 666 (“Section 2254(d)(1) would be undermined if habeas courts introduced rules not clearly established under the guise of extensions to existing law .. . [, although cjertain principles are fundamental enough that when new factual permutations arise, the necessity to apply the earlier rule will be beyond doubt”). And the transposi[*128]tion is improper even on its own terms. According to the Court of Appeals, “Fulminante stands for the proposition that the admission of an additional confession ordinarily reinforces and corroborates the others and is therefore prejudicial.” 574 F. 3d, at 1111. Based on that reading, the Court of Appeals held that the state court’s decision “was contrary to Fulminante.” Id., at 1102. But Fulminante may not be so incorporated into the Strickland performance inquiry.

A state-court adjudication of the performance of counsel under the Sixth Amendment cannot be “contrary to” Fulminante, for Fulminante — which involved the admission of an involuntary confession in violation of the Fifth Amendment — says nothing about the Strickland standard of effectiveness. See Bell v. Cone, 535 U. S. 685, 694 (2002) (“A federal habeas court may issue the writ under the ‘contrary to’ clause if the state court applies a rule different from the governing law set forth in our cases, or if it decides a case differently than we have done on a set of materially indistinguishable facts”). The Fulminante prejudice inquiry presumes a constitutional violation, whereas Strickland seeks to define one. The state court accepted counsel’s view that seeking to suppress Moore’s second confession would have been “fruitless.” It would not have been unreasonable to conclude that counsel could incorporate that view into his assessment of a plea offer, a subject with which Fulminante is in no way concerned.

A finding of constitutionally adequate performance under Strickland cannot be contrary to Fulminante. The state court likely reached the correct result under Strickland. And under § 2254(d), that it reached a reasonable one is sufficient. See Richter, ante, at 109.

B

The Court of Appeals further concluded that it would have been unreasonable for the state postconviction court to have found no prejudice in counsel’s failure to suppress Moore’s[*129] confession to police. To prevail on prejudice before the state court Moore had to demonstrate “a reasonable probability that, but for counsel’s errors, he would not have pleaded guilty and would have insisted on going to trial.” Hill v. Lockhart, 474 U. S. 52, 59 (1985).

Deference to the state court’s prejudice determination is all the more significant in light of the uncertainty inherent in plea negotiations described above: The stakes for defendants are high, and many elect to limit risk by forgoing the right to assert their innocence. A defendant who accepts a plea bargain on counsel’s advice does not necessarily suffer prejudice when his counsel fails to seek suppression of evidence, even if it would be reversible error for the court to admit that evidence.

The state court here reasonably could have determined that Moore would have accepted the plea agreement even if his second confession had been ruled inadmissible. By the time the plea agreement cut short investigation of Moore’s crimes, the State’s case was already formidable and included two witnesses to an admissible confession. Had the prosecution continued to investigate, its case might well have become stronger. At the same time, Moore faced grave punishments. His decision to plead no contest allowed him to avoid a possible sentence of life without parole or death. The bargain counsel struck was thus a favorable one — the statutory minimum for the charged offense — and the decision to forgo a challenge to the confession may have been essential to securing that agreement.

Once again the Court of Appeals reached a contrary conclusion by pointing to Fulminante: “The state court’s finding that a motion to suppress a recorded confession to the police would have been ‘fruitless’... was without question contrary to clearly established federal law as set forth in Fulminante.” 574 F. 3d, at 1112. And again there is no sense in which the state court’s finding could be contrary to Fulminante, for Fulminante says nothing about prejudice for[*130] Strickland purposes, nor does it contemplate prejudice in the plea bargain context.

The Court of Appeals appears to have treated Fulminante as a per se rule of prejudice, or something close to it, in all cases involving suppressible confessions. It is not. In Fulminante five Justices made the uncontroversial observation that many confessions are powerful evidence. See, e. g., 499 U. S., at 296. Fulminante’s prejudice analysis arose on direct review following an acknowledged constitutional error at trial. The State therefore had the burden of showing that it was “clear beyond a reasonable doubt that a rational jury would have found the defendant guilty absent the error.” Neder v. United States, 527 U. S. 1, 18 (1999) (paraphrasing Fulminante, supra). That standard cannot apply to determinations whether inadequate assistance of counsel prejudiced a defendant who entered into a plea agreement. Many defendants reasonably enter plea agreements even though there is a significant probability — much more than a reasonable doubt — that they would be acquitted if they proceeded to trial. Thus, the question in the present case is not whether Moore was sure beyond a reasonable doubt that he would still be convicted if the extra confession were suppressed. It is whether Moore established the reasonable probability that he would not have entered his plea but for his counsel’s deficiency, Hill, supra, at 59, and more to the point, whether a state court’s decision to the contrary would be unreasonable.

To the extent Fulminante’s application of the harmless-error standard sheds any light on the present case, it suggests that the state court’s prejudice determination was reasonable. Fulminante found that an improperly admitted confession was not harmless under Chapman v. California, 386 U. S. 18 (1967), because the remaining evidence against the defendant was weak. The additional evidence consisted primarily of a second confession that Fulminante had made to the informant’s fiance. But many of its details were not[*131] corroborated, the fiance had not reported the confession for a long period of time, the State had indicated that both confessions were essential to its case, and the fiance potentially “had a motive to lie.” 499 U. S., at 300. Moore’s plea agreement, by contrast, ended the government’s investigation well before trial, yet the evidence against Moore was strong. The accounts of Moore’s second confession to his brother and his accomplice’s girlfriend corroborated each other, were given to people without apparent reason to lie, and were reported without delay.

The State gave no indication that its felony-murder prosecution depended on the admission of the police confession, and Moore does not now deny that he kidnaped and killed Rogers. Given all this, an unconstitutional admission of Moore’s confession to police might well have been found harmless even on direct review if Moore had gone to trial after the denial of a suppression motion.

Other than for its discussion of the basic proposition that a confession is often powerful evidence, Fulminante is not relevant to the present case. The state postconviction court reasonably could have concluded that Moore was not prejudiced by counsel’s actions. Under AEDPA, that finding ends federal review. See Richter, ante, at 109.

Judge Berzon’s concurring opinion in the Court of Appeals does not provide a basis for issuance of the writ. The concurring opinion would have found the state court’s prejudice determination unreasonable in light of Kimmelman. It relied on Kimmelman to find that Moore suffered prejudice for Strickland purposes because there was a reasonable possibility that he would have obtained a better plea agreement but for his counsel’s errors. But Kimmelman concerned a conviction following a bench trial, so it did not establish, much less clearly establish, the appropriate standard for prejudice in cases involving plea bargains. See 477 U. S., at 389. That standard was established in Hill, which held that a defendant who enters a plea agreement must show[*132] “a reasonable probability that, but for counsel's errors, he would not have pleaded guilty and would have insisted on going to trial.” 474 U. S., at 59. Moore’s failure to make that showing forecloses relief under AEDPA.

IV

There are certain differences between inadequate-assistance-of-counsel claims in cases, where there was a full trial on the merits and those, like this one, where a plea was entered even before the prosecution decided upon all of the charges. A trial provides the full written record and factual background that serve to limit and clarify some of the choices counsel made. Still, hindsight cannot suffice for relief when counsel’s choices were reasonable and legitimate based on predictions of how the trial would proceed. See Richter, ante, at 108.

Hindsight and second guesses are also inappropriate, and often more so, where a plea has been entered without a full trial or, as in this case, even before the prosecution decided on the charges. The added uncertainty that results when there is no extended, formal record and no actual history to show how the charges have played out at trial works against the party alleging inadequate assistance. Counsel, too, faced that uncertainty. There is a most substantial burden on the claimant to show ineffective assistance. The plea process brings to the criminal justice system a stability and a certainty that must not be undermined by the prospect of collateral challenges in cases not only where witnesses and evidence have disappeared, but also in eases where witnesses and evidence were not presented in the first place. The substantial burden to show ineffective assistance of counsel, the burden the claimant must meet to avoid the plea, has not been met in this case.

The state postconviction court’s decision involved no unreasonable application of Supreme. Court precedent. Because the Court of Appeals erred in finding otherwise, its[*133] judgment is reversed, and the case is remanded for further proceedings consistent with this opinion.

It is so ordered.