Justice Kennedy
delivered the opinion of the Court.
In this case we once again examine when a state prisoner can appeal the denial or dismissal of his petition for writ of habeas corpus. In 1986 two Dallas County assistant district attorneys used peremptory strikes to exclude 10 of the 11 African-Americans eligible to serve on the jury which tried petitioner Thomas Joe Miller-El. During the ensuing 17 years, petitioner has been unsuccessful in establishing, in either state or federal court, that his conviction and death sentence must be vacated because the jury selection procedures violated the Equal Protection Clause and our holding in Batson v. Kentucky, 476 U. S. 79 (1986). The claim now arises in a federal petition for writ of habeas corpus. The procedures and standards applicable in the case are controlled by the habeas corpus statute codified at Title 28, chapter 153, of the United States Code, most recently amended in a substantial manner by the Antiterrorism and Effective Death Penalty Act of 1996 (AEDPA). In the interest of finality AEDPA constrains a federal court’s power to disturb state-court convictions.
The United States District Court for the Northern District of Texas, after reviewing the evidence before the state trial court, determined that petitioner failed to establish a constitutional violation warranting habeas relief. The Court of Appeals for the Fifth Circuit, concluding there was insufficient merit to the case, denied a certificate of appeal-[327]*327ability (COA) from the District Court’s determination. The COA denial is the subject of our decision.
At issue here are the standards AEDPA imposes before a court of appeals may issue a COA to review a denial of habeas relief in the district court. Congress mandates that a prisoner seeking postconviction relief under 28 U. S. C. § 2254 has no automatic right to appeal a district court’s denial or dismissal of the petition. Instead, petitioner must first seek and obtain a COA. In resolving this case we decide again that when a habeas applicant seeks permission to initiate appellate review of the dismissal of his petition, the court of appeals should limit its examination to a threshold inquiry into the underlying merit of his claims. Slack v. McDaniel, 529 U. S. 473, 481 (2000). Consistent with our prior precedent and the text of the habeas corpus statute, we reiterate that a prisoner seeking a COA need only demonstrate “a substantial showing of the denial of a constitutional right.” 28 U. S. C. §2253(c)(2). A petitioner satisfies this standard by demonstrating that jurists of reason could disagree with the district court’s resolution of his constitutional claims or that jurists could conclude the issues presented are adequate to deserve encouragement to proceed further. Slack, supra, at 484. Applying these principles to petitioner’s application, we conclude a COA should have issued.
I
A
Petitioner, his wife Dorothy Miller-El, and one Kenneth Flowers robbed a Holiday Inn in Dallas, Texas. They emptied the cash drawers and ordered two employees, Doug Walker and Donald Hall, to lie on the floor. Walker and Hall were gagged with strips of fabric, and their hands and feet were bound. Petitioner asked Flowers if he was going to kill Walker and Hall. When Flowers hesitated or refused, [328]*328petitioner shot Walker twice in the back and shot Hall in the side. Walker died from his wounds.
The State indicted petitioner for capital murder. He pleaded not guilty, and jury selection took place during five weeks in February and March 1986. When voir dire had been concluded, petitioner moved to strike the jury on the grounds that the prosecution had violated the Equal Protection Clause of the Fourteenth Amendment by excluding African-Americans through the use of peremptory challenges. Petitioner’s trial occurred before our decision in Batson, supra, and Swain v. Alabama, 380 U. S. 202 (1965), was then the controlling precedent. As Swain required, petitioner sought to show that the prosecution’s conduct was part of a larger pattern of discrimination aimed at excluding African-Americans from jury service. In a pretrial hearing held on March 12, 1986, petitioner presented extensive evidence in support of his motion. The trial judge, however, found “no evidence . . . that indicated any systematic exclusion of blacks as a matter of policy by the District Attorney’s office; while it may have been done by individual prosecutors in individual cases.” App. 813. The state court then denied petitioner’s motion to strike the jury. Ibid. Twelve days later, the jury found petitioner guilty; and the trial court sentenced him to death.
Petitioner appealed to the Texas Court of Criminal Appeals. While the appeal was pending, on April 30, 1986, the Court decided Batson v. Kentucky and established its three-part process for evaluating claims that a prosecutor used peremptory challenges in violation of the Equal Protection Clause. First, a defendant must make a prima facie showing that a peremptory challenge has been exercised on the basis of race. 476 U. S., at 96-97. Second, if that showing has been made, the prosecution must offer a race-neutral basis for striking the juror in question. Id., at 97-98. Third, in light of the parties’ submissions, the trial court must deter[329]*329mine whether the defendant has shown purposeful discrimination. Id., at 98.
After acknowledging petitioner had established an inference of purposeful discrimination, the Texas Court of Criminal Appeals remanded the case for new findings in light of Batson. Miller-El v. State, 748 S. W. 2d 459 (1988). A post-trial hearing was held on May 10, 1988 (a little over two years after petitioner’s jury had been empaneled). There, the original trial court admitted all the evidence presented at the Swain hearing and further evidence and testimony from the attorneys in the original trial. App. 843-844.
On January 13, 1989, the trial court concluded that petitioner’s evidence failed to satisfy step one of Batson because it “did not even raise an inference of racial motivation in the use of the state’s peremptory challenges” to support a prima facie ease. App. 876. Notwithstanding this conclusion, the state court determined that the State would have prevailed on steps two and three because the prosecutors had offered credible, race-neutral explanations for each African-American excluded. The court further found “no disparate prosecutorial examination of any of the veniremen in question” and “that the primary reasons for the exercise of the challenges against each of the veniremen in question [was] their reluctance to assess or reservations concerning the imposition of the death penalty.” Id., at 878. There was no discussion of petitioner’s other evidence.
The Texas Court of Criminal Appeals denied petitioner’s appeal, and we denied certiorari. Miller-El v. Texas, 510 U. S. 831 (1993). Petitioner’s state habeas proceedings fared no better, and he was denied relief by the Texas Court of Criminal Appeals.
Petitioner filed a petition for writ of habeas corpus in Federal District Court pursuant to 28 U. S. C. § 2254. Although petitioner raised four issues, we concern ourselves here with only petitioner’s jury selection claim premised on Batson. The Federal Magistrate Judge who considered the merits [330]*330was troubled by some of the evidence adduced in the state-court proceedings. He, nevertheless, recommended, in deference to the state courts’ acceptance of the prosecutors’ race-neutral justifications for striking the potential jurors, that petitioner be denied relief. The United States District Court adopted the recommendation. Pursuant to §2253, petitioner sought a COA from the District Court, and the application was denied. Petitioner renewed his request to the Court of Appeals for the Fifth Circuit, and it also denied the COA.
The Court of Appeals noted that, under controlling habeas principles, a COA will issue “ ‘only if the applicant has made a substantial showing of the denial of a constitutional right.’ ” Miller-El v. Johnson, 261 F. 3d 445, 449 (2001) (quoting 28 U. S. C. § 2253(c)(2)). Citing our decision in Slack v. McDaniel, 529 U. S. 473 (2000), the court reasoned that “[a] petitioner makes a ‘substantial showing’ when he demonstrates that his petition involves issues which are debatable among jurists of reason, that another court could resolve the issues differently, or that the issues are adequate to deserve encouragement to proceed further.” 261 F. 3d, at 449. The Court of Appeals also interjected the requirements of 28 U. S. C. § 2254 into the COA determination: “As an appellate court reviewing a federal habeas petition, we are required by § 2254(d)(2) to presume the state court findings correct unless we determine that the findings result in a decision which is unreasonable in light of the evidence presented. And the unreasonableness, if any, must be established by clear and convincing evidence. See 28 U. S. C. § 2254(e)(1).” 261 F. 3d, at 451.
Applying this framework to petitioner’s COA application, the Court of Appeals concluded “that the state court’s findings are not unreasonable and that Miller-El has failed to present clear and convincing evidence to the contrary.” Id., at 452. As a consequence, the court “determined that the state court’s adjudication neither resulted in a decision that [331]*331was unreasonable in light of the evidence presented nor resulted in a decision contrary to clearly established federal law as determined by the Supreme Court,” ibid,.; and it denied petitioner’s request for a COA. We granted certiorari. 534 U. S. 1122 (2002).
B
While a COA ruling is not the occasion for a ruling on the merit of petitioner’s claim, our determination to reverse the Court of Appeals counsels us to explain in some detail the extensive evidence concerning the jury selection procedures. Petitioner’s evidence falls into two broad categories. First, he presented to the state trial court, at a pretrial Swain hearing, evidence relating to a pattern and practice of race discrimination in the voir dire. Second, two years later, he presented, to the same state court, evidence that directly related to the conduct of the prosecutors in his case. We discuss the latter first.
A comparative analysis of the venire members demonstrates that African-Americans were excluded from petitioner’s jury in a ratio significantly higher than Caucasians were. Of the 108 possible jurors reviewed by the prosecution and defense, 20 were African-American. Nine of them were excused for cause or by agreement of the parties. Of the 11 African-American jurors remaining, however, all but 1 were excluded by peremptory strikes exercised by the prosecutors. On this basis 91% of the eligible black jurors were removed by peremptory strikes. In contrast the prosecutors used their peremptory strikes against just 13% (4 out of 31) of the eligible nonblack prospective jurors qualified to serve on petitioner’s jury.
These numbers, while relevant, are not petitioner’s whole case. During voir dire, the prosecution questioned venire members as to their views concerning the death penalty and their willingness to serve on a capital case. Responses that disclosed reluctance or hesitation to impose capital punishment were cited as a justification for striking a potential [332]*332juror for cause or by peremptory challenge. Wainwright v. Witt, 469 U. S. 412 (1985). The evidence suggests, however, that the manner in which members of the venire were questioned varied by race. To the extent a divergence in responses can be attributed to the racially disparate mode of examination, it is relevant to our inquiry.
Most African-Americans (53%, or 8 out of 15) were first given a detailed description of the mechanics of an execution in Texas:
“[I]f those three [sentencing] questions are answered yes, at some point[,] Thomas Joe Miller-El will be taken to Huntsville, Texas. He will be placed on death row and at some time will be taken to the death house where he will be strapped on a gurney, an IV put into his arm and he will be injected with a substance that will cause his death ... as the result of the verdict in this case if those three questions are answered yes.” App. 215.
Only then were these African-American venire members asked whether they could render a decision leading to a sentence of death. Very few prospective white jurors (6%, or 3 out of 49) were given this preface prior to being asked for their views on capital punishment. Rather, all but three were questioned in vague terms: “Would you share with us . . . your personal feelings, if you could, in your own words how you do feel about the death penalty and capital punishment and secondly, do you feel you could serve on this type of a jury and actually render a decision that would result in the death of the Defendant in this ease based on the evidence? ” Id., at 506.
There was an even more pronounced difference, on the apparent basis of race, in the manner the prosecutors questioned members of the venire about their willingness to impose the minimum sentence for murder. Under Texas law at the time of petitioner’s trial, an unwillingness to do so warranted removal for cause. Huffman v. State, 450 S. W. [333]*3332d 858, 861 (Tex. Crim. App. 1970), vacated in part, 408 U. S. 936 (1972). This strategy normally is used by the defense to weed out pro-state members of the venire, but, ironically, the prosecution employed it here. The prosecutors first identified the statutory minimum sentence of five years’ imprisonment to 34 out of 36 (94%) white venire members, and only then asked: “If you hear a case, to your way of thinking [that] calls for and warrants and justifies five years, you’ll give it?” App. 509. In contrast, only one out of eight (12.5%) African-American prospective jurors were informed of the statutory minimum before being asked what minimum sentence they would impose. The typical questioning of the other seven black jurors was as follows:
“[Prosecutor]: Now, the maximum sentence for [murder] ... is life under the law. Can you give me an idea of just your personal feelings what you feel a minimum sentence should be for the offense of murder the way I’ve set it out for you?
“[Juror]: Well, to me that’s almost like it’s premeditated. But you said they don’t have a premeditated statute here in Texas.
“[Prosecutor]: Again, we’re not talking about self-defense or accident or insanity or killing in the heat of passion or anything like that. We’re talking about the knowing—
“[Juror]: I know you said the minimum. The minimum amount that I would say would be at least twenty years.” Id., at 226-227.
Furthermore, petitioner points to the prosecution’s use of a Texas criminal procedure practice known as jury shuffling. This practice permits parties to rearrange the order in which members of the venire are examined so as to increase the likelihood that visually preferable venire members will be moved forward and empaneled. With no information about [334]*334the prospective jurors other than their appearance, the party requesting the procedure literally shuffles the juror cards, and the venire members are then reseated in the new order. Tex. Code Crim. Proc. Ann., Art. 35.11 (Vernon Supp. 2003). Shuffling affects jury composition because any prospective jurors not questioned during voir dire are dismissed at the end of the week, and a new panel of jurors appears the following week. So jurors who are shuffled to the back of the panel are less likely to be questioned or to serve.
On at least two occasions the prosecution requested shuffles when there were a predominant number of African-Americans in the front of the panel. On yet another occasion the prosecutors complained about the purported inadequacy of the card shuffle by a defense lawyer but lodged a formal objection only after the postshuffle panel composition revealed that African-American prospective jurors had been moved forward.
Next, we turn to the pattern and practice evidence adduced at petitioner’s pretrial Swain hearing. Petitioner subpoenaed a number of current and former Dallas County assistant district attorneys, judges, and others who had observed firsthand the prosecution’s conduct during jury selection over a number of years. Although most of the witnesses denied the existence of a systematic policy to exclude African-Americans, others disagreed. A Dallas County district judge testified that, when he had served in the District Attorney’s Office from the late-1950’s to early-1960’s, his superior warned him that he would be fired if he permitted any African-Americans to serve on a jury. Similarly, another Dallas County district judge and former assistant district attorney from 1976 to 1978 testified that he believed the office had a systematic policy of excluding African-Americans from juries.
Of more importance, the defense presented evidence that the District Attorney’s Office had adopted a formal policy to exclude minorities from jury service. A 1963 circular by the [335]*335District Attorney’s Office instructed its prosecutors to exercise peremptory strikes against minorities: '“Do not take Jews, Negroes, Dagos, Mexicans or a member of any minority race on a jury, no matter how rich or how well educated.’ ” App. 710. A manual entitled “Jury Selection in a Criminal Case” was distributed to prosecutors. It contained an article authored by a former prosecutor (and later a judge) under the direction of his superiors in the District Attorney’s Office, outlining the reasoning for excluding minorities from jury service. Although the manual was written in 1968, it remained in circulation until 1976, if not later, and was available at least to one of the prosecutors in Miller-El’s trial. Id., at 749, 774, 783.
Some testimony casts doubt on the State’s claim that these practices had been discontinued before petitioner’s trial. For example, a judge testified that, in 1985, he had to exclude a prosecutor from trying cases in his courtroom for race-based discrimination in jury selection. Other testimony indicated that the State, by its own admission, once requested a jury shuffle in order to reduce the number of African-Americans in the venire. Id., at 788. Concerns over the exclusion of African-Americans by the District Attorney’s Office were echoed by Dallas County’s Chief Public Defender.
This evidence had been presented by petitioner, in support of his Batson claim, to the state and federal courts that denied him relief. It is against this background that we examine whether petitioner’s ease should be heard by the Court of Appeals.
II
As mandated by federal statute, a state prisoner seeking a writ of habeas corpus has no absolute entitlement to appeal a district court’s denial of his petition. 28 U. S. C. §2253. Before an appeal may be entertained, a prisoner who was denied habeas relief in the district court must first seek and [336]*336obtain a COA from a circuit justice or judge. This is a jurisdictional prerequisite because the COA statute mandates that “[u]nless a circuit justice or judge issues a certificate of appealability, an appeal may not be taken to the court of appeals. . . § 2253(c)(1). As a result, until a COA has been issued federal courts of appeals lack jurisdiction to rule on the merits of appeals from habeas petitioners.
A COA will issue only if the requirements of § 2253 have been satisfied. “The COA statute establishes procedural rules and requires a threshold inquiry into whether the circuit court may entertain an appeal.” Slack, 529 U. S., at 482; Hohn v. United States, 524 U. S. 236, 248 (1998). As the Court of Appeals observed in this case, § 2253(c) permits the issuance of a COA only where a petitioner has made a “substantial showing of the denial of a constitutional right.” In Slack, supra, at 483, we recognized that Congress codified our standard, announced in Barefoot v. Estelle, 463 U. S. 880 (1983), for determining what constitutes the requisite showing. Under the controlling standard, a petitioner must “sho[wj that reasonable jurists could debate whether (or, for that matter, agree that) the petition should have been resolved in a different manner or that the issues presented were ‘adequate to deserve encouragement to proceed further.’” 529 U. S., at 484 (quoting Barefoot, supra, at 893, n. 4).
The COA determination under § 2253(c) requires an overview of the claims in the habeas petition and a general assessment of their merits. We look to the District Court’s application of AEDPA to petitioner’s constitutional claims and ask whether that resolution was debatable amongst jurists of reason. This, threshold inquiry does not require full consideration of the factual or legal bases adduced in support of the claims. In fact, the statute forbids it. When a court of appeals sidesteps this process by first deciding the merits of an appeal, and then justifying its denial of a COA based [337]*337on its adjudication of the actual merits, it is in essence deciding an appeal without jurisdiction.
To that end, our opinion in Slack held that a COA does not require a showing that the appeal will succeed. Accordingly, a court of appeals should not decline the application for a COA merely because it believes the applicant will not demonstrate an entitlement to relief. The holding in Slack would mean very little if appellate review were denied because the prisoner did not convince a judge, or, for that matter, three judges, that he or she would prevail. It is consistent with §2253 that a COA will issue in some instances where there is no certainty of ultimate relief. After all, when a COA is sought, the whole premise is that the prisoner “ ‘has already failed in that endeavor.’ ” Barefoot, supra, at 893, n. 4.
Our holding should not be misconstrued as directing that a COA always must issue: Statutes such as AEDPA have placed more, rather than fewer, restrictions on the power of federal courts to grant writs of habeas corpus to state prisoners. Duncan v. Walker, 533 U. S. 167, 178 (2001) (“ ‘AEDPA’s purpose [is] to further the principles of comity, finality, and federalism’” (quoting Williams v. Taylor, 529 U. S. 420, 436 (2000))); Williams v. Taylor, 529 U. S. 362, 399 (2000) (opinion of O’Connor, J.). The concept of a threshold, or gateway, test was not the innovation of AEDPA. Congress established a threshold prerequisite to appealability in 1908, in large part because it was “concerned with the increasing number of frivolous habeas corpus petitions challenging capital sentences which delayed execution pending completion of the appellate process .. ..” Barefoot, supra, at 892, n. 3. By enacting AEDPA, using the specific standards the Court had elaborated earlier for the threshold test, Congress confirmed the necessity and the requirement of differential treatment for those appeals deserving of attention from those that plainly do not. It follows that issuance of a COA must not be proforma or a matter of course.
[338]*338A prisoner seeking a COA must prove “ ‘something more than the absence of frivolity’ ” or the existence of mere “good faith” on his or her part. Barefoot, supra, at 893. We do not require petitioner to prove, before the issuance of a COA, that some jurists would grant the petition for habeas corpus. Indeed, a claim can be debatable even though every jurist of reason might agree, after the COA has been granted and the case has received full consideration, that petitioner will not prevail. As we stated in Slack, “[w]here a district court has rejected the constitutional claims on the merits, the showing required to satisfy § 2253(c) is straightforward: The petitioner must demonstrate that reasonable jurists would find the district court’s assessment of the constitutional claims debatable or wrong.” 529 U. S., at 484.
Since Miller-El’s claim rests on a Batson violation, resolution of his COA application requires a preliminary, though not definitive, consideration of the three-step framework mandated by Batson and reaffirmed in our later precedents. E. g., Purkett v. Elem, 514 U. S. 765 (1995) (per curiam); Hernandez v. New York, 500 U. S. 352 (1991) (plurality opinion). Contrary to the state trial court’s ruling on remand, the State now concedes that petitioner, Miller-El, satisfied step one: “[T]here is no dispute that Miller-El presented a prima facie claim” that prosecutors used their peremptory challenges to exclude venire members on the basis of race. Brief for Respondent 32. Petitioner, for his part, acknowledges that the State proceeded through step two by proffering facially race-neutral explanations for these strikes. Under Batson, then, the question remaining is step three: whether Miller-El “has carried his burden of proving purposeful discrimination.” Hernandez, supra, at 359.
As we confirmed in Purkett v. Elem, 514 U. S., at 768, the critical question in determining whether a prisoner has proved purposeful discrimination at step three is the persua[339]*339siveness of the prosecutor’s justification for his peremptory strike. At this stage, “implausible or fantastic justifications may (and probably will) be found to be pretexts for purposeful discrimination.” Ibid. In that instance the issue comes down to whether the trial court finds the prosecutor’s race-neutral explanations to be credible. Credibility can be measured by, among other factors, the prosecutor’s demeanor; by how reasonable, or how improbable, the explanations are; and by whether the proffered rationale has some basis in accepted trial strategy.
In Hernandez v. New York, a plurality of the Court concluded that a state court’s finding of the absence of discriminatory intent is “a pure issue of fact” accorded significant deference:
“Deference to trial court findings on the issue of discriminatory intent makes particular sense in this context because, as we noted in Batson, the finding ‘largely will turn on evaluation of credibility.’ 476 U. S., at 98, n. 21. In the typical peremptory challenge inquiry, the decisive question will be whether counsel’s race-neutral explanation for a peremptory challenge should be believed. There will seldom be much evidence bearing on that issue, and the best evidence often will be the demeanor of the attorney who exercises the challenge. As with the state of mind of a juror, evaluation of the prosecutor’s state of mind based on demeanor and credibility lies ‘peculiarly within a trial judge’s province.’ Wainwright v. Witt, 469 U. S. 412, 428 (1985), citing Patton v. Yount, 467 U. S. 1025, 1038 (1984).” 500 U. S., at 365.
Deference is necessary because a reviewing court, which analyzes only the transcripts from voir dire, is not as well positioned as the trial court is to make credibility determinations. “[I]f an appellate court accepts a trial court’s finding that a prosecutor’s race-neutral explanation for his peremptory challenges should be believed, we fail to see how the [340]*340appellate court nevertheless could find discrimination. The credibility of the prosecutor’s explanation goes to the heart of the equal protection analysis, and once that has been settled, there seems nothing left to review.” Id., at 367.
In the context of direct review, therefore, we have noted that “the trial court’s decision on the ultimate question of discriminatory intent represents a finding of fact of the sort accorded great deference on appeal” and will not be overturned unless clearly erroneous. Id., at 364. A federal court’s collateral review of a state-court decision must be consistent with the respect due state courts in our federal system. Where 28 U. S. C. §2254 applies, our habeas jurisprudence embodies this deference. Factual determinations by state courts are presumed correct absent clear and convincing evidence to the contrary, § 2254(e)(1), and a decision adjudicated on the merits in a state court and based on a factual determination will not be overturned on factual grounds unless objectively unreasonable in light of the evidence presented in the state-court proceeding, § 2254(d)(2); see also Williams, 529 U. S., at 399 (opinion of O’Connor, J.).
Even in the context of federal habeas, deference does not imply abandonment or abdication of judicial review. Deference does not by definition preclude relief. A federal court can disagree with a state court’s credibility determination and, when guided by AEDPA, conclude the decision was unreasonable or that the factual premise was incorrect by clear and convincing evidence. In the context of the threshold examination in this Batson claim the issuance of a COA can be supported by any evidence demonstrating that, despite the neutral explanation of the prosecution, the peremptory strikes in the final analysis were race based. It goes without saying that this includes the facts and circumstances that were adduced in support of the prima facie case. Cf. Reeves v. Sanderson Plumbing Products, Inc., 530 U. S. 133 (2000) (in action under Title VII of the Civil Rights Act of 1964, employee’s prima facie case and evidence that employer’s [341]*341race-neutral response was a pretext can support a finding of purposeful discrimination). Only after a COA is granted will a reviewing court determine whether the trial court’s determination of the prosecutor’s neutrality with respect to race was objectively unreasonable and has been rebutted by clear and convincing evidence to the contrary. At this stage, however, we only ask whether the District Court’s application of AEDPA deference, as stated in §§ 2254(d)(2) and (e)(1), to petitioner’s Batson claim was debatable amongst jurists of reason.
C
Applying these rules to Miller-El’s application, we have no difficulty concluding that a COA should have issued. We conclude, on our review of the record at this stage, that the District Court did not give full consideration to the substantial evidence petitioner put forth in support of the prima facie case. Instead, it accepted without question the state court’s evaluation of the demeanor of the prosecutors and jurors in petitioner’s trial. The Court of Appeals evaluated Miller-El’s application for a COA in the same way. In ruling that petitioner’s claim lacked sufficient merit to justify appellate proceedings, the Court of Appeals recited the requirements for granting a writ under § 2254, which it interpreted as requiring petitioner to prove that the state-court decision was objectively unreasonable by clear and convincing evidence.
This was too demanding a standard on more than one level. It was incorrect for the Court of Appeals, when looking at the merits, to merge the independent requirements of §§ 2254(d)(2) and (e)(1). AEDPA does not require petitioner to prove that a decision is objectively unreasonable by clear and convincing evidence. The clear and convincing evidence standard is found in § 2254(e)(1), but that subsection pertains only to state-court determinations of factual issues, rather than decisions. Subsection (d)(2) contains the unreasonable [342]*342requirement and applies to the granting of habeas relief rather than to the granting of a COA.
The Court of Appeals, moreover, was incorrect for an even more fundamental reason. Before the issuance of a COA, the Court of Appeals had no jurisdiction to resolve the merits of petitioner’s constitutional claims. True, to the extent that the merits of this case will turn on the agreement or disagreement with a state-court factual finding, the clear and convincing evidence and objective unreasonableness standards will apply. At the COA stage, however, a court need not make a definitive inquiry into this matter. As we have said, a COA determination is a separate proceeding, one distinct from the underlying merits. Slack, 529 U. S., at 481; Hohn, 524 U. S., at 241. The Court of Appeals should have inquired whether a “substantial showing of the denial of a constitutional right” had been proved. Deciding the substance of an appeal in what should only be a threshold inquiry undermines the concept of a COA. The question is the debatability of the underlying constitutional claim, not the resolution of that debate.
In this case, the statistical evidence alone raises some debate as to whether the prosecution acted with a race-based reason when striking prospective jurors. The prosecutors used their peremptory strikes to exclude 91% of the eligible African-American venire members, and only one served on petitioner’s jury. In total, 10 of the prosecutors’ 14 peremptory strikes were used against African-Americans. Happenstance is unlikely to produce this disparity.
The case for debatability is not weakened when we examine the State’s defense of the disparate treatment. The Court of Appeals held that “[t]he presumption of correctness is especially strong, where, as here, the trial court and state habeas court are one and the same.” 261 F. 3d, at 449. As we have noted, the trial court held its Batson hearing two years after the voir dire. While the prosecutors had proffered contemporaneous race-neutral justifications for many [343]*343of their peremptory strikes, the state trial court, had no occasion to judge the credibility of these explanations at that time because our equal protection jurisprudence then, dictated by Swain, did not require it. As a result, the evidence presented to the trial court at the Batson hearing was subject to the usual risks of imprecision and distortion from the passage of time.
In this case, three of the State’s proffered race-neutral rationales for striking African-American jurors pertained just as well to some white jurors who were not challenged and who did serve on the jury. The prosecutors explained that their peremptory challenges against six African-American potential jurors were based on ambivalence about the death penalty; hesitancy to vote to execute defendants capable of being rehabilitated; and the jurors’ own family history of criminality. In rebuttal of the prosecution’s explanation, petitioner identified two empaneled white jurors who expressed ambivalence about the death penalty in a manner similar to their African-American counterparts who were the subject of prosecutorial peremptory challenges. One indicated that capital punishment was not appropriate for a first offense, and another stated that it would be “difficult” to impose a death sentence. Similarly, two white jurors expressed hesitation in sentencing to death a defendant who might be rehabilitated; and four white jurors had family members with criminal histories. As a consequence, even though the prosecution’s reasons for striking African-American members of the venire appear race neutral, the application of these rationales to the venire might have been selective and based on racial considerations. Whether a comparative juror analysis would demonstrate the prosecutors’ rationales to have been pretexts for discrimination is an unnecessary determination at this stage, but the evidence does make debatable the District Court’s conclusion that no purposeful discrimination occurred.
[344]*344We question the Court of Appeals’ and state trial court’s dismissive and strained interpretation of petitioner’s evidence of disparate questioning. 261 F. 3d, at 452 (“The findings of the state court that there was no disparate questioning of the Batson jurors . . . [is] fully supported by the record”). Petitioner argues that the prosecutors’ sole purpose in using disparate questioning was to elicit responses from the African-American venire members that reflected an opposition to the death penalty or an unwillingness to impose a minimum sentence, either of which justified for-cause challenges by the prosecution under the then-applicable state law. This is more than a remote possibility. Disparate questioning did occur. Petitioner submits that disparate questioning created the appearance of divergent opinions even though the venire members’ views on the relevant subject might have been the same. It follows that, if the use of disparate questioning is determined by race at the outset, it is likely a justification for a strike based on the resulting divergent views would be pretextual. In this context the differences in the questions posed by the prosecutors are some evidence of purposeful discrimination. Batson, 476 U. S., at 97 (“Similarly, the prosecutor’s questions and statements during voir dire examination and in exercising his challenges may support or refute an inference of discriminatory purpose”).
As a preface to questions about views the prospective jurors held on the death penalty, the prosecution in some instances gave an explicit account of the execution process. Of those prospective jurors who were asked their views on capital punishment, the preface was used for 53% of the African-Americans questioned on the issue but for just 6% of white persons. The State explains the disparity by asserting that a disproportionate number of African-American venire members expressed doubts as to the death penalty on their juror questionnaires. This cannot be accepted without further inquiry, however, for the State’s own evidence is in[345]*345consistent with that explanation. By the State’s calculations, 10 African-American and 10 white prospective jurors expressed some hesitation about the death penalty on their questionnaires; however, of that group, 7 out of 10 African-Americans and only 2 out of 10 whites were given the explicit description.
There is an even greater disparity along racial lines when we consider disparate questioning concerning minimum punishments. Ninety-four percent of whites were informed of the statutory minimum sentence, compared to only twelve and a half percent of African-Americans. No explanation is proffered for the statistical disparity. Pierre v. Louisiana, 306 U. S. 354, 361-362 (1939) (“ ‘The fact that the testimony ... was not challenged by evidence appropriately direct, cannot be brushed aside.’ Bad there been evidence obtainable to contradict and disprove the testimony offered by petitioner, it cannot be assumed that the State would have refrained from introducing it” (quoting Norris v. Alabama, 294 U. S. 587, 594-595 (1935))). Indeed, while petitioner’s appeal was pending before the Texas Court of Criminal Appeals, that court found a Batson violation where this precise line of disparate questioning on mandatory mínimums was employed by one of the same prosecutors who tried the instant case. Chambers v. State, 784 S. W. 2d 29, 31 (Tex. Crim. App. 1989). It follows, in our view, that a fair interpretation of the record on this threshold examination in the COA analysis is that the prosecutors designed their questions to elicit responses that would justify the removal of African-Americans from the venire. Batson, supra, at 93 (“Circumstantial evidence of invidious intent may include proof of disproportionate impact.. .. We have observed that under some circumstances proof of discriminatory impact ‘may for all practical purposes demonstrate unconstitutionality because in various circumstances the discrimination is very difficult to explain on nonracial grounds’ ”).
[346]*346We agree with petitioner that the prosecution’s decision to seek a jury shuffle when a predominant number of African-Americans were seated in the front of the panel, along with its decision to delay a formal objection to the defense’s shuffle until after the new racial composition was revealed, raise a suspicion that the State sought to exclude African-Americans from the jury. Our concerns are amplified by the fact that the state court also had before it, and apparently ignored, testimony demonstrating that the Dallas County District Attorney’s Office had, by its own admission, used this process to manipulate the racial composition of the jury in the past. App. 788 (noting that a prosecutor admitted to requesting a jury shuffle “because a predominant number of the first six, eight or ten jurors were blacks”). Even though the practice of jury shuffling might not be denominated as a Batson claim because it does not involve a peremptory challenge, the use of the practice here tends to erode the credibility of the prosecution’s assertion that race was not a motivating factor in the jury selection.
Finally, in our threshold examination, we accord some weight to petitioner’s historical evidence of racial discrimination by the District Attorney’s Office. Evidence presented at the Swain hearing indicates that African-Americans almost categorically were excluded from jury service. Batson, supra, at 94 (“Proof of systematic exclusion from the venire raises an inference of purposeful discrimination because the ‘result bespeaks discrimination’ ”); Vasquez v. Hillery, 474 U. S. 254, 259 (1986) (“As early as 1942, this Court rejected a contention that absence of blacks on the grand jury was insufficient to support an inference of discrimination, summarily asserting that ‘chance or accident could hardly have accounted for the continuous omission of negroes from the grand jury lists for so long a period as sixteen years or more’ ” (quoting Hill v. Texas, 316 U. S. 400, 404 (1942))); Hernandez v. Texas, 347 U. S. 475, 482 (1954) (“But it taxes our credulity to say that mere chance resulted in there being [347]*347no members of this class among the over six thousand jurors called in the past 25 years”). Only the Federal Magistrate Judge addressed the import of this evidence in the context of a Batson claim; and he found it both unexplained and disturbing. Irrespective of whether the evidence could prove sufficient to support a charge of systematic exclusion of African-Americans, it reveals that the culture of the District Attorney’s Office in the past was suffused with bias against African-Americans in jury selection. This evidence, of course, is relevant to the extent it casts doubt on the legitimacy of the motives underlying the State’s actions in petitioner’s case. Even if we presume at this stage that the prosecutors in Miller-El’s case were not part of this culture of discrimination, the evidence suggests they were likely not ignorant of it. Both prosecutors joined the District Attorney’s Office when assistant district attorneys received formal training in excluding minorities from juries. The supposition that race was a factor could be reinforced by the fact that the prosecutors marked the race of each prospective juror on their juror cards.
In resolving the equal protection claim against petitioner, the state courts made no mention of either the jury shuffle or the historical record of purposeful discrimination. We adhere to the proposition that a state court need not make detailed findings addressing all the evidence before it. This failure, however, does not diminish its significance. Our concerns here are heightened by the fact that, when presented with this evidence, the state trial court somehow reasoned that there was not even the inference of discrimination to support a prima facie case. This was clear error, and the State declines to defend this particular ruling. “If these general assertions were accepted as rebutting a defendant’s prima facie case, the Equal Protection Clause ‘would be but a vain and illusory requirement.’ ” Batson, 476 U. S., at 98 (quoting Norris, 294 U. S., at 598).
[348]*348To secure habeas relief, petitioner must demonstrate that a state court’s finding of the absence of purposeful discrimination was incorrect by clear and convincing evidence, 28 U. S. C. § 2254(e)(1), and that the corresponding factual determination was “objectively unreasonable” in light of the record before the court. The State represents to us that petitioner will not be able to satisfy his burden. That may or may not be the case. It is not, however, the question before us. The COA inquiry asks only if the District Court’s decision was debatable. Our threshold examination convinces us that it was.
The judgment of the Fifth Circuit is reversed, and the case is remanded for further proceedings consistent with this opinion.
It is so ordered.