Self report measures of smoking may contain substantial measurement error. If these errors are nonrandom, then they may be correlated with substantive variables in epidemiologic or intervention studies, thus confounding and biasing estimates of structural relations. In efforts to circumvent such bias, investigators have supplemented questionnaires with biological indicators of exposure to tobacco smoke. However, errors in biological indicators may also include systematic errors that produce biased estimates. The current study was designed to estimate the variance of systematic and random errors in self report and two biologic measures of smoking in adolescents. A linear model was fit to data from 342 males and 250 females with repeated measures separated by 1 year. Thiocyanate and self report were found to have substantial nonrandom error components that were correlated with psychosocial variables. In contrast, errors in carbon monoxide in expired air were not autocorrelated, and were uncorrelated with psychosocial variables.