The comparability of self-report and observer measures of substance abuse among 118 homeless mentally ill persons was assessed using cross-sectional and longitudinal measures. Possible correlates of nondisclosure were identified from demographic variables and clinical indicators. Lifetime abuse reported at baseline was a sensitive predictor of subsequent abuse behavior in the project, but cross-sectional measures based only on self-report or observer ratings failed to identify many abusers. A total of 17% of the subjects never disclosed abuse that was observed during the project. The level of substance abuse is likely to be severely underestimated among homeless mentally ill persons when only one self-report measure is used at just one point in time. This problem can, however, largely be overcome by incorporating information from observers and from multiple follow-ups or by focusing on lifetime rather than current abuse. We also conclude that underreporting may bias estimates of some correlates of substance abuse. (Authors)