Anthony George Iron Cloud, Sr. was convicted*, by a jury of involuntary manslaughter under 18 U.S.C. §§ 1153 and 1112. He appeals the district court’s decision to admit into evidence the results of a portable breath test (“PBT”) claiming that its admission deprived him of a fair trial. We reverse and remand for a new trial.
I. FACTS
At 9:30 p.m. on January 19, 1997, Anthony Iron Cloud gave a friend a ride to his home in the Evergreen housing area of Pine Ridge Reservation. According to Iron Cloud, he had consumed six beers over the course of six hours.2 When Iron Cloud reached the housing area, he saw a pedestrian, Abel Iron Rope, walking along the right-hand side of the road. According to Iron Cloud, he pulled over to the left to avoid the pedestrian. He then turned to look at his passenger and hit Iron Rope. It is undisputed that Iron Cloud never applied his brakes and was driving 15 miles per hour in a 10 müe per hour zone before the accident occurred.
Abel Iron Rope had an extensive history of mental illness and was characterized by his family as a danger to himself and to the community. Iron Rope also had a history of jumping in front of cars. In 1993, he jumped in front of a propane truek on the highway causing the truck to careen into a ditch. Furthermore, a police officer testified that at approximately 4:00 p.m. on January 19, 1997, the day of the accident, Iron Rope ran out directly in front of his patrol car and the officer was barely able to avoid hitting him.
After the accident, Iron Cloud and his friend checked on Iron Rope, who was sitting up and nodded when asked if he was fine. Iron Cloud left the scene of the accident, went home, and asked his wife to call 911.3 The police arrived shortly thereafter. Twenty-five minutes after the accident, Tribal Officer Twiss administered a portable breath test (“PBT”) which indicated a blood-alcohol level of .14 percent. Iron Cloud was arrested and approximately one hour after the accident he was given an intoxilyzer breath test which registered his blood-alcohol level as. 11 percent. Five hours after the crash, a blood test was taken and registered .033 percent. Abel Iron Rope subsequently died as a result of the accident.
Iron Cloud was charged with involuntary manslaughter under 18 U.S .C. §§ 1153 and 1112 for operating his motor vehicle in a grossly negligent manner. Iron Cloud filed a motion in limine to exclude the PBT as evidence of anything more than a screening test used to determine probable cause for arrest. The district court overruled the motion without an evidentiary hearing. At trial, the government’s witness was allowed to calculate a blood-alcohol level reading based on a hypothetical question which included the PBT and the other tests performed on *590Iron Cloud. The expert concluded that Iron Cloud’s blood-alcohol level at the time of the accident was .13 or .14 percent.4 At the conclusion of the trial, the district court gave a limiting jury instruction which provided that the PBT should not be considered in isolation, but could be considered with the other tests as proof of intoxication. The jury found Iron Cloud guilty and he was sentenced to 21 months imprisonment and three years supervised release. Iron Cloud appeals the admission of the PBT as substantive proof of his intoxication.
II. DISCUSSION
Admissibility of the PBT
In United States v. Black Cloud, 101 F.3d 1258 (8th Cir.1996), this court set forth a two part test to determine when scientific testimony is admissible. First, the district court must “determine whether the testimony is based on reliable scientific technique, and whether it will assist the jury.” Black Cloud, 101 F.3d at 1261 (citation omitted). In its determination, the district court should consider the following factors when assessing the reliability of a scientific technique: “(1) whether the technique can be and has been tested; (2) whether the technique has been subjected to peer review and publication; (3) the known or potential rate of error for the technique and the existence and maintenance of standards for controlling the technique’s operation; and (4) whether the technique is generally accepted in the scientific community.” Id. (citing Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579, 593-95, 113 S.Ct. 2786, 125 L.Ed.2d 469 (1993)). Second, even if the district court concludes that the expert testimony is admissible, it “may exclude the testimony if the testimony has an unfairly prejudicial effect that substantially outweighs its probative value.” Id. (citations omitted).
The district court in this case did not follow either of these steps. The court refused to hold a Daubert hearing to determine the reliability of the test, stating before the jury: “[the] PBT test does not require a Daubert hearing. The PBT test is a test which has been recognized in the scientific community and as such, Daubert is not appropriate.” Tr. at 43. Neither did the court analyze whether the test would be unfairly prejudicial. We review a district court’s decision to admit evidence for an abuse of discretion. Black Cloud, 101 F.3d at 1261.
By denying Iron Cloud’s request for a Daubert hearing on the reliability of the PBT, the judge took the accuracy of the PBTs for granted and he ignored established procedure. Contrary to the court’s statements concerning the reliability of the PBT, the PBT has not been established as reliable. PBTs are used in the field for screening purposes. The government’s expert witness, Roger Mathison, agreed that the PBT was only a preliminary screening test. Tr. at 197-199. Furthermore, almost every state that has addressed the issue has refused to admit the results of the test for purposes other than probable cause.5 Although the admissibility of evi*591dence is governed by federal standards,6 in the face of this overwhelming case law as to the limited reliability of the PBT, we conclude, without further foundation being laid, that the PBT is not reliable as anything more than a screening test to be used for probable cause.
We are not persuaded by the government’s argument that the district court was correct in admitting the results of the PBT test without a Daubert hearing because the technology has been in use for an extended period of time. The mere fact that a test has been used for a long time does not make it reliable. The same argument could be made for polygraph tests, which clearly are not admissible in this circuit. See United States v. Williams, 95 F.3d 723 (8th Cir.1996); Conner v. Auger, 595 F.2d 407 (8th Cir.1979).
Finally, we hold the error was not cured by the judge’s limiting jury instruction. Jury Instruction No. 12 stated as follows:
The Court has permitted the introduction of evidence of a portable breath test (PBT) taken of the defendant. You are instructed that a PBT test is not recognized, standing alone, as proof of intoxication. It may be considered, however, together with the tests of the blood, breath (intoxilyzer) and other evidence pointing to the question of whether the defendant was under the influence of alcohol at the time of the event for which the defendant stands charged.
As we have discussed, the PBT should not be used for anything other than probable cause determinations. This instruction simply makes matters worse by instructing the jury that they can consider the PBT as substantive evidence, albeit in conjunction with other evidence.
Harmless Error
This court has held that “[evidentiary rule violations ‘which do not affect [a defendant’s] constitutional rights are subject to Fed.R.Crim.P. 52(a) harmless error analysis.’ ” United States v. DeAngelo, 13 F.3d 1228, 1233 (8th Cir.1994). Under this court’s interpretation of Fed.R.Crim.P. 52(a), “[a]n error is harmless if the reviewing court, after reviewing the entire record, determines that no substantial rights of the defendant were affected, and that the error did not influence or had only a very slight influence on the verdict.” United States v. Wilcox, 50 F.3d 600, 603 (8th Cir.1995) (citation omitted). We will reverse the conviction “only if the jury may have been substantially swayed by improperly-admitted evidence....” Id. (citations omitted).
Iron Cloud was accused of involuntary manslaughter under 18 U.S.C. § 1153 and 1112. The indictment was based upon *592the allegation that Iron Cloud operated his motor vehicle in a “grossly negligent manner” with actual knowledge that his conduct was a threat to the lives of others or with actual knowledge that would reasonably enable him to foresee the danger to others. The government sought to prove this allegation primarily by showing that Iron Cloud was intoxicated, and heavily relied on the PBT results to do so. We find that the jury may have been improperly influenced by this admitted evidence.
The admission of the PBT results becomes significant considering the government’s primary contentions. The government used the PBT results in two ways. First, the government used the results to establish that Iron Cloud was intoxicated. Without the PBT results, Iron Cloud might have successfully disputed the level of his intoxication. The PBT was taken only 25 minutes after the accident and registered a blood-alcohol level of .14 percent, well over the legal limit. Without that level, as the defense expert testified, based on the blood-alcohol reading of Iron Cloud taken some five hours after the accident, Iron Cloud’s blood-alcohol level at the time of the accident could have been argued to be between .7 and .8 percent, which is under the legal limit. In giving this opinion, Dr. Robert Looyenga, a chemist who was employed for twenty-five years at South Dakota School of Mines & Technology, assumed the rate of elimination was somewhere between .011 and .015.7
The experts disagreed on the reliability of the intoxilyzer test. The defense counsel challenged the accuracy of both the PBT and the intoxilyzer tests. The intoxilyzer test is measured from the use of the breath, not the blood, so a ratio must be applied to get the results in the form of the blood-alcohol level. The defense contended that this ratio can lead to inaccuracies. The government’s expert conceded on cross-examination that the blood test was the most reliable of the three tests and that the intoxilyzer test brought many variables into play.8 Of course, the government disputed the challenges to the intoxilyzer’s accuracy.9 It was obviously for the jury to determine the credibility of the conflicting opinions of the two experts. Our primary concern, however, is whether the admission of the PBT test, which showed that Iron Cloud had a much higher alcohol content, bolstered the reliability of the intoxilyzer test and, therefore, influenced the jury in ac-*593eepting the testimony of the government expert over the testimony of the defense. We cannot be sure that the PBT did not substantially sway the jury in this manner.
The admission of the PBT to prove intoxication is also important considering the circumstances surrounding the accident. It is undisputed that Abel Iron Rope had mental instabilities and had run out in front of cars on at least two previous occasions. In fact, he ran in front of a police officer’s car earlier in the afternoon on the day of the accident in question and was nearly hit. Without the PBT to support the theory that Iron Cloud was intoxicated, it is possible the jury could have found that the accident occurred because Abel Iron Rope ran in front of Iron Cloud’s car.
Second, the government used the PBT results to undermine the veracity of the defendant. Iron Cloud testified that he drank only six beers that day. According to the government’s expert’s testimony, however, Iron Cloud would have had to consume at least eleven beers within that time period to have a blood-alcohol level of .13 or .14, the level derived by considering the PBT. Without the PBT results, Iron Cloud’s blood-alcohol level could have been as low as .07 or .08 at the time of the accident. As a result, Iron Cloud’s testimony would have been consistent with the evidence of the lower blood-alcohol level and his credibility would not have been so easily challenged.
We cannot say that the jury was not greatly influenced by the high blood-alcohol level registered through the PBT shortly after the accident, either as proof of the defendant’s intoxication or of his lack of credibility. Therefore, admitting the PBT results was not harmless error.
III. CONCLUSION
For the foregoing reasons, the judgment of the district court is reversed and the case is remanded to the district court for a new trial.
. The exact amount of alcohol that Iron Cloud had consumed is a matter of dispute. Iron Cloud contends that he had six beers. Based upon the various tests given to Iron Cloud after the accident, including the PBT, the government contended that Iron Cloud drank eleven beers during that time period. There was also disputed evidence that Iron Cloud had been smoking marijuana. This informa-lion was obtained by a blood test taken five hours after the accident. The government’s expert testified that Iron Cloud smoked marijuana within the time period of one hour before the crash to when the blood was taken. Iron Cloud contends that he had not smoked marijuana for two days before the accident.
.Iron Cloud claims that he left the scene because he was afraid of being attacked by bystanders and went straight home to call 911.
. Without using the PBT evidence, Iron Cloud’s expert testified that his blood-alcohol level would have been .07 to .09 percent. The government contends that even without the PBT his blood-alcohol level would still be over .10 percent.
. See also Boyd v. City of Montgomery, 472 So.2d 694, 697 (Ala.Crim.App.1985) (holding that preliminary breath tests are only admissible to establish probable cause); Patrick v. State, 295 Ark. 473, 750 S.W.2d 391, 394 (1988) (stating that PBT's are admissible only on behalf of the defendant because they are unreliable); Attix v. Voshell, 579 A.2d 1125, 1129 (Del.Super.Ct.1989) (holding that the PBT can be admitted only for probable cause and not for substantial evidence because no court has established that it is reliable); State v. Zell, 491 N.W.2d 196, 197 (Iowa Ct.App.1992) (stating, "[t]he results of the preliminary screening test are inadmissible because the test is inherently unreliable and may register an inaccurate percentage of alcohol present in the breath, and may also be inaccurate as to the presence or absence of any alcohol at all”); People v. Keskinen, 177 Mich. *591App. 312, 441 N.W.2d 79, 82 (1989) (holding that court erred in admitting evidence of the defendant's preliminary breath test); Justice v. Director of Revenue, 890 S.W.2d 728, 731 (Mo.Ct.App.1995) (stating that PBTs are not admissible by statute); State v. Strizich, 286 Mont. 1, 952 P.2d 1365, 1371 (1997) (holding that the PBT is intended only for determining. probable cause); State v. Klingelhoefer, 222 Neb. 219, 382 N.W.2d 366, 369-70 (1986) (holding that preliminary test is only relevant for limited purpose of establishing probable cause); City of Fargo v. Ruetker, 490 N.W.2d 481, 482-83 (N.D.1992) (holding that an alcohol screening test cannot be admitted if a defendant admits probable cause); Commonwealth v. Stanley, 427 Pa.Super. 422, 629 A.2d 940, 942 (1993) (stating that PBT results are inadmissible); Jones v. Town of Marion, 28 Va.App. 791, 508 S.E.2d 921 (1999) (citing to slate statute which provides that preliminary breath tests are only to be used in determining probable cause); Thompson v. State Dept. of Licensing, 91 Wash.App. 887, 960 P.2d 475, 477 (1998) (holding that "the l'e-sults of a portable breath test are not admissible as evidence at trial or to establish probable cause for arrest”); State v. Beaver, 181 Wis.2d 959, 512 N.W.2d 254, 258-59 (1994) (PBT not admitted). Compare State v. Huettl, 379 N.W.2d 298, 305 (S.D.1985) (holding that PBT results were inadmissible because of state implied consent statues); State v. Anderson, 359 N.W.2d 887 (S.D.1984) (holding that because the PBT is a field sobriety test for establishing probable cause, the results are not admissible against a defendant.).
. United States v. McMillan, 820 F.2d 251, 255 (8th Cir.1987).
. There was a disagreement between the experts as to the proper rate of elimination to be used in this case. The experts agree that the average person's rate is .015. Dr. Looyenga stated that the vast majority of people have rates of elimination between .01 and .02, and it is just as accurate to use the lower rate as it is to use the upper rate in computing a blood-alcohol level. The government's expert, Mr. Mathison, however, used the elimination rate of .019 to arrive at a blood-alcohol level of .13 at the time of the accident. This rate of elimination was determined using the results from the intoxilyzer and the blood test taken five hours after the accident. On cross-examination, Mr. Mathison admitted that if the elimination rate was determined using the PBT and the intoxilyzer, it would be .06, which is an unrealistic rate.
. Dr. Mathison testified:
Q. Okay. Again, because of the variance, because you are not actually taking blood and measuring it, now you've got another variance in there. You are trying to transfer information taken from a breath test into a blood content and you have to apply an arbitrary ratio, arbitrary only in the sense that it varies individual to individual?
A. Yes, that’s correct.
Q. There can be, again, a range of inaccuracy up to 10 percent on that, isn’t that right?
A. That’s correct.
Q. In addition to that, if the machine is not calibrated properly and regularly, then the intoxilyzer test can be off because of erroneous calibration?
A. Certainly is possible with the use of the intoxilyzer the calibration is checked before and after the actual breath test is—
Tr. at 202-03.
.As we pointed out in McMillan, the variables associated with the intoxilyzer breath test go to the weight and not the admissibility of the tests. McMillan, 820 F.2d at 255.