Given that prosecutors often willingly and even eagerly use highly suspect testimony by police and informants, especially jailhouse informants and accomplices who testify under plea agreements, the polygraph could be used to restore public confidence in testimony by police officers and cooperating witnesses and to prevent miscarriages of justice. The author argues that polygraph results should not be admitted as evidence in their own right, but rather should be used as a tool to screen out untrustworthy evidence. Courts should exclude testimony from a witness who has tested "deceptive" when that result is corroborated by a "nondeceptive" result on the opposing side ("paired results"). Using only paired results resolves the unquantifiable uncertainties and reduces by a factor of at least five the quantifiable uncertainties that underlie most legitimate resistance to the polygraph. If individual examination results are incorrect 20 percent of the time, the chance that two together will be in error is only 4 percent. Evidence that has at most a 4 percent likelihood of being true is too untrustworthy to warrant submitting it for a jury's consideration. Excluding it will reduce the incidence of perjury and the number of erroneous verdicts. It will discourage frivolous claims and frivolous defenses. When possible, the decision to exclude testimony should be made before trial. This article addresses broad problems of admissibility and exclusion of evidence that affect both civil and criminal cases. In the instance of criminal cases, the author considers the constitutional implications of the recommended approach, particularly those having to do with the right against self-incrimination. It also analyzes alternative approaches and their drawbacks.