The authors surveyed a cohort of 1,184 adolescents in the 7th, 8th, and 9th grades. Measures of tobacco, alcohol, and marijuana use and of constructs from 3 theoretical models of substance use were obtained at each point. Clustering analysis for 3-wave substance use data indicated subgroups of nonusers, minimal experimenters, late starters, and escalators. Discriminant function analyses tested whether study variables differentiated the subgroups. One discriminant function accounted for the majority of between-group association; it had loadings for (high) life stress, nonadaptive coping, deviance-prone attitudes, and parental and peer substance use, and (low) parental support, academic competence, and behavioral control. Escalators were high on this function; late starters and experimenters had intermediate values; and nonusers were low on the function. Implications for theories of vulnerability to substance abuse are discussed.