BEAM, Circuit Judge.

Wayne Dumond appeals from the magistrate's denial of his petition for habeas corpus relief. We affirm, in part, reverse, in part, and remand for further proceedings.

I. BACKGROUND

On the afternoon of September 11, 1984, a seventeen-year-old high school student in Forrest City, Arkansas, was abducted at gunpoint from her home. The male assailant drove the victim to a secluded area in her automobile and parked the car in a location hidden from view. At that point, the man and the victim exited the car. From there, the victim was led down a short path and forced to lie down.

The assailant removed the victim's jeans and underpants and placed them beneath her. He then forced her to engage in vaginal intercourse. Her aggressor next forced her to perform fellatio. The assailant then again engaged in vaginal intercourse.

After the rape was committed, the victim was threatened with death but ultimately was allowed to dress, and the assailant returned the victim to her home. The victim showered and changed her clothes. She reported the rape and accompanied the police to the scene of the attack where evidence was found to confirm her account of the crime. The victim described the assailant as tall, really thin, with dishwater-blonde hair, crystal blue eyes, and a full beard which was darker than his hair.

During a photographic show-up, the victim indicated that Ricky White resembled the assailant. However, White was working in another part of the state on the day of the rape, and she did not identify him as the rapist at a one-person lineup. Later, Walter Stevenson, who matched the assailant's description and worked near a restaurant which the victim frequented, was placed in a lineup. She did not identify Stevenson as her assailant. Woodcutters working near the area of her home on the date of the rape were also brought in for lineups but none were recognized by the victim. On approximately October 29, 1984, the victim observed Dumond driving a pick-up truck on a Forrest City street and immediately identified him as the perpetrator of the crime. Dumond was taken into custody, placed in a lineup, and identified by her as the man who kidnapped and raped her. At the time of the arrest, the official police report described Dumond as being six-feet tall and weighing one hundred forty-five pounds, with brown hair and hazel eyes.

At his trial, Dumond contended that he was ill at home when the rape occurred. Several family members and friends testified, corroborating Dumond's claim. Dumond was found guilty of kidnap and rape and received consecutive sentences of life imprisonment and twenty years. Dumond's conviction was affirmed by the Arkansas Supreme Court, Dumond v. State, 290 Ark. 595, 721 S.W.2d 663 (1986) ( Dumond I), and his application for post-conviction relief was denied, Dumond v. State, 294 Ark. 379, 743 S.W.2d 779 (1988) ( Dumond II). Dumond then filed a petition in the district court seeking habeas corpus relief. Pursuant to 28 U.S.C. § 636(c)(1) (1982), the parties consented to the jurisdiction of a United States Magistrate. The magistrate denied Dumond relief. This appeal followed.

For reversal, Dumond argues that the magistrate erred by quashing subpoenas, that he was denied due process because an immunoglobulin allotype test exculpated him, that he was denied effective assistance of counsel in a number of particulars, and that he was denied due process because the state withheld evidence exonerating him.

II. DISCUSSION [10] A. Immunoglobulin Allotyping

The victim testified at trial that the assailant wore a prophylactic during the initial act of vaginal intercourse and did not ejaculate. She further stated that her aggressor ejaculated during oral sex and that she expectorated the semen. According to her testimony, the man again engaged in vaginal intercourse for "about three seconds" after forcing her to perform oral sex but did not then ejaculate. Thus, she testified that her attacker only ejaculated once and that it occurred during the oral-sex act.

At trial, Charles Dorsey, a serologist from the Arkansas State Crime Laboratory, testified regarding an ABO blood-grouping test he performed on the semen samples taken from the victim's clothing. During cross-examination, Dorsey concluded that twenty-eight percent of the population, which included Dumond, could have produced the semen. In 1987, Dumond submitted the clothing to Dr. Moses Schanfield, an expert in genetic testing, and requested that Dr. Schanfield conduct an immunoglobulin allotype test. Dr. Schanfield had genetic allotyping performed on the semen found on the victim's pant leg. Schanfield concluded that based on the test, there was a ninety-nine plus percent probability that Dumond was not the rapist because the semen lacked a genetic marker which Dumond possessed. However, Dr. Schanfield's conclusion was based on the assumption that vaginal fluids were not mixed with the semen used for the test. If the semen was intermixed with vaginal secretions, Dr. Schanfield reported that the results would be inconclusive.

Based on this newly discovered evidence, Dumond sought post-conviction relief pursuant to Ark.R.Crim.P. 37.1. Pointing to the victim's testimony that a single ejaculation occurred during oral sex, Dumond asserted that the semen was pure and he could not, therefore, be the rapist. Dumond contended that due to the genetic allotype evidence, he was entitled to a new trial as a matter of due process. However, the Arkansas Supreme Court determined that "[a] claim of new evidence is a direct rather than a collateral attack on the judgment and not within the purview of our post-conviction rule." Dumond II, 294 Ark. at 385, 743 S.W.2d at 782 (citation omitted). Dumond renewed this claim in his petition for a writ of habeas corpus and sought to compel the testimony of the victim at the hearing on the habeas petition. The magistrate quashed the subpoena and determined that the record did not support Dumond's contention that the semen on the pant leg was pure. Dumond contends that the magistrate erred in discounting the victim's testimony regarding the number and location of ejaculations. Dumond further argues that to reach the decision, the magistrate was required to speculate concerning factual issues upon which the victim could have testified and which have never been litigated. We agree.

It is well-established that "the existence merely of newly discovered evidence relevant to the guilt of a state prisoner is not a ground for relief on federal habeas corpus." Townsend v. Sain, 372 U.S. 293, 317, 83 S.Ct. 745, 759, 9 L.Ed.2d 770 (1963). To be the basis for relief, the "evidence must bear upon the constitutionality of the applicant's detention." Id. Interpreting this standard, we have concluded that relief will not be granted unless it can be shown that the evidence "would probably produce an acquittal on retrial." Mastrian v. McManus, 554 F.2d 813, 823 (8th Cir.) (citations omitted), cert. denied, 433 U.S. 913, 97 S.Ct. 2985, 53 L.Ed.2d 1099 (1977).

We are troubled by the inconsistencies in the evidence. As indicated, the victim testified that the assailant ejaculated once and that it occurred during oral sex. Yet, as Dr. Schanfield stated, "The physical evidence based on Mr. Dorsey's analysis and confirmed by our testing indicates that there [were] vaginal ejaculation and that there [were] large amounts of semen deposited vaginally." Habeas record at 173. Attempting to solve this problem, the magistrate presented a number of scenarios under which the semen and vaginal fluids may have been combined. If the victim had testified at the hearing on the petition, the inconsistency may have been resolved. It seems a more prudent course to allow her to testify before sailing on a sea of speculation. Significantly, questions concerning the number of ejaculations and the location of the semen have never been adequately addressed in any proceeding.

At the time of the trial, these questions lacked the importance they now possess in light of the immunoglobulin allotyping. Therefore, the victim was asked only if she knew "what [the semen] may have gotten on." Trial record at 573. She simply replied, "No, sir." Id. Moreover, the Arkansas Supreme Court refused to consider the issue when it was raised in the post-conviction proceeding, precluding an evidentiary hearing on the issue. We are not unmindful that an inquiry into these issues may resurrect painful memories. At the same time, we must balance this possibility against Dumond's rights. If the semen used for the genetic allotyping was pure, then an innocent man may be imprisoned. This we cannot chance. However, without her testimony on this issue, we are incapable of determining whether the newly discovered evidence "would probably produce an acquittal on retrial." We, therefore, remand to the magistrate to conduct a hearing on the genetic allotype evidence. At the hearing, counsel may examine the victim concerning the facts relevant to the results of the immunoglobulin allotyping. Dumond's and the state's expert witnesses on genetic testing may also be examined at the proceeding.

B. The Lineup

After the trial, Henry Leary, a St. Francis County Deputy Sheriff, stated in an affidavit that the victim did not initially identify Dumond when she viewed the lineup in which Dumond participated; that the police chief escorted her to another room; and that upon her return, she identified Dumond without hesitation. The inference is that she was "coached" to identify Dumond during the brief period in which she left the viewing room.

Dumond asserts that the victim's in-court identification was tainted by the improper lineup. Dumond argues that the state suppressed the evidence of lineup misconduct in violation of Brady v. Maryland, 373 U.S. 83, 83 S.Ct. 1194, 10 L.Ed.2d 215 (1963). In Brady, the Court held that evidence that is both favorable to the accused and material to either guilt or punishment must be disclosed. Id. at 87, 83 S.Ct. at 1196. Presumably, if Dumond had been aware of this evidence, he would have moved to suppress the in-court identification or, failing that, used the evidence to impeach the victim's testimony.

Because the Leary affidavit was newly discovered evidence, the Arkansas Supreme Court determined, as it had with the immunoglobulin allotype evidence, that the affidavit was "a direct attack on the conviction and [was] not therefore cognizable under our postconviction rule." Dumond II, 294 Ark. at 394, 743 S.W.2d at 787 (citation omitted). Though the magistrate reached the merits of the issue, he rejected Dumond's Brady claim for several reasons. The magistrate determined that Leary's version of the events surrounding the lineup was not credible. As an alternative basis for his decision, the magistrate gave deference to the Arkansas Supreme Court's finding that the in-court identification was based on the attack, not on the lineup. The magistrate, therefore, concluded that even had the purported evidence of lineup misconduct been disclosed, the result of Dumond's trial would not have been different; thus, the evidence was not material. Dumond first contends that the magistrate had no basis for disbelieving Leary's affidavit.

"The [magistrate] is in the best position to observe the demeanor of the witnesses and to assess credibility, and such factual findings shall not be set aside unless clearly erroneous." United States v. Risken, 869 F.2d 1098, 1100 (8th Cir. 1989) (citing Anderson v. City of Bessemer City, 470 U.S. 564, 573-74, 105 S.Ct. 1504, 1511-12, 84 L.Ed.2d 518 (1985); Fed.R.Civ.P. 52(a)). Officers Stacy Dye and James Cole conducted the lineup. At the hearing on the habeas petition, Dye and Cole testified that the victim immediately identified Dumond and that she was not prompted to choose Dumond. Joe Goff, the Forrest City Chief of Police, stated at the post-conviction hearing that he was not present at the lineup. We agree with the magistrate that it is difficult to believe that the police would need to prod the victim to identify Dumond when it was her prior identification which led the police to place Dumond in the lineup. Considering the conflicting testimony regarding the lineup, we cannot say that the magistrate's finding here was clearly erroneous. Having determined under our narrow standard of review that no lineup wrongdoing occurred, we need not reach Dumond's other claims regarding identification. Accordingly, we reject Dumond's Brady claim.

III. CONCLUSION

We have carefully examined all other issues and find them to be without merit. For the foregoing reasons, we reverse and remand this matter to the magistrate to conduct a hearing consistent with this opinion.