Self-reports of drug use frequency are central to treatment outcome evaluations, estimates of the prevalence of heavy use, estimates of treatment need, and other questions with direct relevance to drug policies. Nevertheless, surprisingly little is known about the validity of these self-reports. This study examines the accuracy of 701 frequency self-reports made by a sample of methadone maintenance clients. Self-report accuracy is evaluated by comparing rates of positive urinalyses found for each case with rates that would be expected had drug use occurred only as often as reported. Expected rates of positive urinalyses were derived from conservative Monte Carlo models of drug use for each case. This procedure revealed extensive heroin and cocaine use frequency underreporting. After adjusting for frequency underreporting, 51% of 279 cases reporting only occasional heroin use (1-10 days in the past 30), and 22% of the 157 cases reporting occasional cocaine use, were found to be using these drugs with frequencies corresponding to what the Office of National Drug Control Policy defines as "hardcore use" (more than 10 days in the past 30). Drug use frequency underreporting appears substantial, and might constitute an important threat to the validity of some treatment outcome evaluations, needs assessments and other analyses that rely on drug use frequency self-reports.