Psychiatric and psychological diagnoses are imperfect. Unlike somatic medicine, most psychological and psychiatric phenomena have no gold standard to establish their presence beyond reasonable doubt. Consequently, prevalence estimates are based on the average agreement of imperfect evaluators. Küchenhoff, Augustin, and Kunz (2012) provided a statistical method for estimating confidence intervals of the prevalence based on the well-known kappa coefficient of interrater agreement. We expand this method and derive confidence intervals for the probability of a diagnosis being true (i.e., the positive predictive value). We illustrate the method and its results with empirical data for a particular type of paraphilia (pedophilia) in sexual offenders. The findings indicate that up to 1 in 3 diagnoses of pedophilia may be wrong. Given the similar rates of prevalence and interrater agreement reported for diagnoses in general psychiatry (such as schizophrenia or affective disorders), the results likely apply to other diagnostic domains as well.