Nanon Williams appeals the denial of his habeas petition under 28 U.S.C. § 2254 (2006), following his capital murder conviction in Texas state court. For the reasons set forth below, we REVERSE and REMAND to the district court for a full evidentiary hearing on the ineffective assistance of counsel claims on which the Certificate of Appealability was granted.
I. FACTUAL BACKGROUND
A. The Murder Trial
In 1995, a jury convicted Williams of capital murder and sentenced him to *354death.1 The facts underlying Williams’s conviction are disputed, but we can discern a basic outline of those facts from the state court record. On May 13, 1992, Williams, his friend Vaal Guevara, and Guevara’s friend Elaine Winn, went to Adonius Collier’s apartment to arrange a drug transaction. Collier, his friend Ammade Rasul, and Rasul’s girlfriend, Stephanie Anderson, met with Williams, Guevara, and Winn. They agreed to complete the transaction at a nearby park. The participants all went to the park in several vehicles. The Government alleged at trial that Williams carried a .25-caliber pistol and a shotgun hidden in his clothing while Guevara carried a .22-caliber pistol. Upon arriving at the park, Williams, Guevara, Rasul, and Collier went into the woods to conduct the drug transaction. The remaining participants stayed in the cars.
During the course of the drug transaction, gunfire erupted. Rasul testified that Williams shot him once in the face with the pistol and that he suffered a bullet wound in the foot as he ran toward the parking lot. Forensic testimony proffered at trial linked the bullet from Rasul’s foot to Williams’s .25-caliber pistol. Rasul and Anderson sought medical attention for Ra-sul’s injuries, and Anderson reported the shooting to a Houston police officer. Thereafter, the police investigated the park and found Collier’s dead body. Collier had suffered a shotgun wound to the head. The medical examiners who performed Collier’s autopsy recovered some shotgun pellets from Collier’s cranial cavity and a spent and mutilated bullet, which was marked as “EB-1.”2 Winn, Anderson, and another person, identified at trial only as “Xavier,” were present at the park, but the only potential eyewitnesses to the shooting were Collier (the deceased), Ra-sul, Guevara, and Williams. Williams did not testify. Thus, the only eyewitness testimony at trial came from Rasul and Guevara.
Rasul testified that after Williams shot him, he ran. While running, Rasul heard more shots including, apparently, the shot to his foot. Because Rasul ran before Collier was shot, while Guevara stayed behind, Guevera’s story and its credibility were very important. Guevara initially testified that he did not fire his gun at all. Later, Guevara testified that he fired his gun in the direction of Collier but did not strike him. Guevara then testified that he ran after Rasul. He stated that he never actually saw Williams shoot Collier, but he heard shotgun fire and saw Collier’s feet twitching. Guevara also said that he heard Williams say “no more witnesses” before hearing the shotgun fire.
The “objective” evidence in the case consisted of expert testimony about the ammunition recovered from Collier’s head and the cause of Collier’s death. The assistant medical examiner for Harris County, Dr. Brown, performed the autopsy. Dr. Brown testified that the shotgun pellets killed Collier and that Collier was still alive when shot with the shotgun. He based this conclusion on the fact that there was a red margin around Collier’s bullet wound. According to Dr. Brown, this red margin indicated blood pressure at the time the wound was inflicted. Dr. Brown did not find any evidence of “EB-1” during *355the autopsy and could not account for its presence, but he admitted that Collier could have been shot with a bullet before the shotgun blast and that the shotgun blast could have then obliterated evidence of a prior bullet wound.
The other expert was a Houston police department criminalist, Robert Baldwin, who specifically testified that the “EB-1” bullet came from a .25-caliber pistol like the one witnesses claimed Williams had carried and not from a .22-caliber pistol like the one Guevara admitted to carrying. Baldwin admitted that he failed to test fire the pistols, but testified unequivocally that his analysis was correct.
Williams’s trial counsel, Loretta Mul-drow, never sought an independent ballistics test or an independent autopsy or examination of the pathology report. Even after learning that Baldwin had not test fired the pistols in his examination of the ballistics evidence, Muldrow still did not request a continuance to have an independent expert further examine this evidence. Consequently, Baldwin’s unre-butted testimony buttressed and gave credibility to Guevara’s testimony that he was not Collier’s killer. Indeed, throughout the State’s case, notably in closing argument, the prosecutor made much of the fact that the jury did not need to decide which shot killed Collier because the objective evidence showed that all shots came from firearms attributable to Williams, not to Guevara.
B. The State Habeas Proceeding
On direct appeal, Williams’s conviction and sentence were affirmed. In conneetion with the habeas proceedings, the state trial court ordered the release of Guevara’s .22-caliber pistol and the ballistics evidence. Before turning that evidence over, the Harris County District Attorney’s Office retested the ballistics evidence. In so doing, it determined that “EB-1” had, in fact, been fired by Guevara’s .22-caliber pistol in direct contradiction to Baldwin’s trial testimony. Williams used this information as a foundation for several claims in his habeas proceeding. Because the Certificate of Appealability was granted only on the ineffective assistance of counsel claims,3 we focus on the evidence only as it relates to those issues.
In the state habeas proceeding, Williams presented Baldwin’s recantation of his trial testimony. Baldwin admitted that the “EB-1” bullet came from Guevara’s gun. To complement Baldwin’s recantation, Williams also presented an affidavit from Ronald Singer, Chief Criminologist with the Tarrant County Medical Examiner’s Office, whom Williams retained to independently test the ballistics evidence. Singer opined that Collier had suffered two wounds to the head, one inflicted by a .22-caliber weapon and one inflicted by a shotgun. According to Singer, the shotgun wound could have obliterated the smaller wound if the smaller wound had been inflicted first. Singer further stated that the bullet recovered from Rasul’s foot had not been fired from the same weapon as “EB-1.” He opined that, even in its damaged state, “EB-1” and the .25-caliber bullet *356taken from Rasul’s foot were easily distinguishable, particularly with the aid of a comparison microscope. Singer believed that a competent examiner should have had no problem discerning the difference and that Baldwin’s trial testimony “at best demonstrates extreme carelessness on his part and at worst calls into question his expertise.” In addition to Baldwin’s recantation and Singer’s testimony, Williams presented affidavits from two jurors stating that they would have voted differently had they known of the correct ballistics evidence.
Williams also submitted an affidavit from Dr. Andrew Krouse, Deputy Chief Medical Examiner in Tarrant County, whom he hired to examine the autopsy evidence and trial testimony. Dr. Krouse indicated that the pistol wound (from Guevara’s gun) alone carried a high probability of fatality for Collier. He disagreed with Dr. Brown’s trial testimony about the significance of the red margin around Collier’s bullet wound. Dr. Krouse testified that it was impossible to state with certainty the order in which the two firearm wounds were inflicted or that the shotgun wound was the cause of death. While Dr. Krouse believed that both wounds should have been listed as causes of death, he also believed that there was a likelihood that “EB-1” was the cause of Collier’s death.
Finally, Williams submitted trial counsel Muldrow’s affidavit indicating that she should have sought funding for independent experts, particularly a ballistics expert. She opined that had she done so, she would have found a “winning” defense.
The state trial court granted an eviden-tiary hearing. Singer testified consistently with his affidavit. Muldrow indicated that she had viewed the theory that some one else shot Collier and killed him as a viable defense, but one unsupported by evidence at trial. She admitted that her decision not to seek independent ballistics evidence was not a strategic one, but rather one based on the faulty assumption that the state had done an adequate job. Mul-drow conceded that, even with accurate ballistics evidence, she would have had to account for the shotgun blast and the evidence placing the shotgun in Williams’s hand.4 Nonetheless, she considered her conduct deficient.
After the first evidentiary hearing, the original trial judge left the bench. A second evidentiary hearing was held before the new judge in which substantially the same testimony was presented except that Muldrow did not testify again.
The state trial court entered findings of fact and conclusions of law recommending that habeas relief be granted only on the claim that counsel was ineffective in failing to retain a firearms expert to show that “EB-1” came from Guevara’s gun. In reaching this conclusion, the trial court specifically found that “EB-1” came from Collier’s head and was fired from Guevara’s gun. The trial court simultaneously found that Williams had not demonstrated that “EB-1” inflicted the fatal wound or that Collier was already dead when Williams shot him with the shotgun. Instead, the trial court found that Collier *357was alive when Williams shot him with the shotgun. As a result, the trial court recommended denial of Williams’s claim of actual innocence.
On the ineffective assistance claims, however, the trial court found that proper ballistics testing would have caused counsel to seek out independent pathological testimony on the possibility that “EB-1,” not the shotgun, caused Collier’s death. This evidence, in turn, would have changed the type and strength of cross-examination of Guevara, the jury’s ultimate assessment of Guevara’s credibility, and much of the prosecution’s closing argument. The trial court further determined that the jury would have heard and possibly credited Krouse’s testimony that Collier was dead before the shotgun blast. The trial court concluded that, but for counsel’s deficiencies, the jury would have been presented with favorable and arguably exculpatory evidence and might have found Williams guilty of a lesser offense. Accordingly, there was a reasonable probability of a different result sufficient enough to satisfy the prejudice prong of Strickland v. Washington, 466 U.S. 668, 695, 104 S.Ct. 2052, 80 L.Ed.2d 674 (1984).
Despite the trial court’s recommendation, the Texas Court of Criminal Appeals denied habeas relief on all grounds. The appellate court entered a two-page order denying relief. The court offered no explanation or analysis to support its decision other than the following:
Evidentiary hearings were held and the trial judge entered findings of fact and conclusions of law as to the allegations raised. In those findings and conclusions the trial court has recommended relief be granted due to ineffective assistance of counsel. However, we do not believe, based on our review of the record presented, that some of the crucial fact findings and the recommendation based, at least in part, on them, are supported by the evidence presented at the evidentiary hearing.
Accordingly, after a review of the record with respect to the allegations made by Applicant, and with due consideration of the recommendation made by the trial court, the relief sought is denied.
C. The Federal Habeas Proceeding
Following the decision of the Texas Court of Criminal Appeals, Williams filed a § 2254 petition in the United States District Court for the Southern District of Texas. Among other claims, Williams raised a claim of ineffective assistance of counsel based on trial counsel’s failure to obtain independent ballistics evidence and the testimony of an independent pathologist. Williams also asserted that the cumulative effect of trial counsel’s ineffectiveness and the falseness of suppressed evidence undermined confidence in the jury verdict. The district court determined that it should not conduct an evi-dentiary hearing. Instead, it granted summary judgment to the state on all claims relating to Williams’s conviction.5
With respect to the ineffective assistance of counsel claims relevant here, the district court faced the difficult issue of how to handle the Texas Court of Criminal Appeals’s decision regarding the state habeas court’s factual findings. Under the Antiterrorism and Effective Death Penalty Act (AEDPA), deference must be given to factual findings of the state court in the absence of clear and convincing evidence to the contrary. 28 U.S.C. § 2254(e)(1). Because the ineffective assistance of counsel claims were adjudicat*358ed in state court, the district court, as well as our court, must defer to the state court’s resolution unless its determination was “contrary to” or an “unreasonable application of’ clearly established federal law as determined by the United States Supreme Court. § 2254(d); see also Hill v. Johnson, 210 F.3d 481, 485 (2000). A state court decision is contrary to clearly established federal law if it “applies a rule that contradicts the governing law set forth in [Supreme Court] eases,” Williams v. Taylor, 529 U.S. 362, 405, 120 S.Ct. 1495, 146 L.Ed.2d 389 (2000), or “if the state court confronts a set of facts that are materially indistinguishable from a decision of [the Supreme Court] and nevertheless arrives at a result different from [the Court’s] precedent.” Id. at 406, 120 S.Ct. 1495. A state court decision involves an unreasonable application of clearly established federal law if the state court “correctly identifies the governing legal rule but applies it unreasonably to the facts of a particular prisoner’s case .... ” Id. at 407-08, 120 S.Ct. 1495.
Because the Texas Court of Criminal Appeals failed to specify which of the state trial court’s factual findings it rejected, the district court diligently and conscientiously attempted to discern on its own which factual findings survived and which did not. The district court, in a very thorough and deliberate opinion, ultimately concluded that only the factual findings inconsistent with the denial of habeas relief — the state trial court’s findings regarding the prejudicial effect of counsel’s alleged errors — would be disregarded. It concluded that the remaining trial court factual findings were entitled to deference and denied habeas relief. This appeal followed.
II. DISCUSSION
On appeal, Williams challenges the district court’s approach because the Texas Court of Criminal Appeals did not explain which facts it accepted and which it rejected. He contends that the district court erroneously substituted its own judgment, guessing which facts were accepted and rejected, and incorrectly applied a presumption of correctness to the facts it determined were accepted.
A presumption of correctness applies to explicit findings of fact and to “those unarticulated findings which are necessary to the state court’s conclusions of mixed law and fact.” Valdez v. Cockrell, 274 F.3d 941, 948 n. 11 (5th Cir.2001). A trial court’s factual findings are entitled to a presumption of correctness even if the state appellate court reached a different legal conclusion when applying the law to those facts. Sumner v. Mata, 449 U.S. 539, 546 — 47, 101 S.Ct. 764, 66 L.Ed.2d 722 (1981); Craker v. Procunier, 756 F.2d 1212, 1213-14 (5th Cir.1985). On the other hand, a state habeas trial court’s factual findings do not survive review by the Texas Court of Criminal Appeals where they were neither adopted nor incorporated into the appellate court’s peremptory denial of relief, but instead were directly inconsistent with the appellate court’s decision. Micheaux v. Collins, 944 F.2d 231, 232 (5th Cir.1991) (en banc). In Micheaux, this Court held that, in such a circumstance, the presumption of correctness did not attach to the state habeas trial court’s findings and the federal district court correctly conducted a de novo hearing on the petitioner’s claims. Id.; see also Singleton v. Johnson, 178 F.3d 381, 384-85 (5th Cir. 1999) (concluding that when the state ha-beas trial court recommends that relief be granted in part and denied in part but the Texas Court of Criminal Appeals denies relief without a written order, there are no factual findings to which the presumption of correctness attaches, requiring the fed*359eral district court to conduct a de novo evidentiary hearing).
Given the Texas Court of Criminal Appeals’s decision, it would be pure speculation for this court or the district court to determine which facts the Texas Court of Criminal Appeals concluded were and were not supported by the evidence. Accordingly, we conclude that this case is analogous to Micheaux and Singleton, not Craker, such that no factual findings survive to which we can give deference. As a result, the district court must conduct a de novo hearing of Williams’s ineffective assistance of counsel claims premised on trial counsel’s failings with respect to the ballistics and pathological evidence, as well as Williams’s cumulative error point.
III. CONCLUSION
We REVERSE the trial court’s grant of summary judgment to the State on the question of Williams’s conviction, and REMAND for a full de novo evidentiary hearing of Williams’s ineffective assistance of counsel claims on which the Certificate of Appealability was granted.