Research involving drug users and treatment evaluations continue to rely extensively on self-reports of drug use. This paper presents a meta-analytical review of 24 studies published since 1985 that examined the validity of drug self-reports in high risk populations. Only studies employing a biological criterion of validity (e.g., urinalysis, hair analysis) are included. Coefficients of chance-corrected agreement between self-reports and the validity criteria are calculated from published data to facilitate cross-study comparisons. The median conditional kappa (kappa c) was .42, considerably below the level of kappa c = .80 that represents acceptable reporting accuracy. The magnitude of drug use underreporting documented in this review could seriously bias prevalence estimates and treatment outcome studies.