First, the equivalence of measurement models for four demographic groups was examined separately for 12 risk factors and two measures of substance use. Then these 14 measures were correlated with five external criteria to assess measurement validity. Results imply that one measurement model fits the data for 11 of the 12 risk factors. For one risk factor (peer drug modeling) and the two drug-use scales, observed group differences in the factor loadings stemmed from differences in the distributions of a few drug-related items that were infrequently endorsed, especially by black females. No meaningful group differences in the validity coefficients that related the scales to external criteria were observed. The results are optimistic for the study of race and gender differences in the correlates of substance use. Research on group differences in the prediction of rare behaviors should examine group differences in distributions that may affect measurement differently for different groups. 7 tables, 1 figures, and 43 references