Longitudinal surveys of adolescent drug use rely heavily on self-report measures. The reliability of such measures are therefore of concern to researchers. As a result, a considerable amount of empirical research has sought to identify sources of adolescents' inaccurate responses to questions about personal drug use. One way to do this is to examine the consistency of adolescent drug-use self-reports over time. Inconsistencies in adolescents' reporting of drug use over time may result from a variety of causes, including lying, errors in recall, and random errors. In an attempt to identify factors related to inconsistencies in self-reported drug use, this study used a multilevel analysis in a longitudinal cohort study with four waves of data to identify significant predictors of denying ever having used a substance after having admitted using it at an earlier wave. The dataset was collected as part of the Television, School, and Family Project, a longitudinal study of adolescent smoking prevention in Los Angeles and San Diego. This project used a longitudinal cohort design with randomization of schools to various treatment conditions. A cohort of 6,695 students from 287 classrooms within 47 public schools was surveyed with self-administered questionnaires at four time points. The study found that male adolescents had significantly higher rates of inconsistent responses over time than did female adolescents. Black and Hispanic adolescents had significantly higher rates of inconsistent responses regarding ever having used alcohol and cigarettes (only for black adolescents) than white adolescents. The subjects' living status and academic achievements were significant predictors of inconsistent responses regarding having ever used marijuana. Thus, these results are consistent with the notion that inconsistent responses may bias the estimation of the prevalence of having ever used drugs in multivariate analyses. 6 tables and 39 references