While both macro- and micro-level studies have identified important correlates of substance use, multi-level models may explain more than each level alone. Drawing on extant research and Akers' (1998) Social Structure-Social Learning model, we offer hypotheses about the relationship between contextual- and individual-level explanatory variables and substance use. The sample included 85,000 students in 202 school districts. Hierarchical linear modeling revealed low socioeconomic status, percent rural, and racial composition directly affected use of some substances net individuals' characteristics. Further, the effects of gender, age, and class were substantially mediated by differential association and school bonds, which significantly influenced alcohol, marijuana, ecstasy, and methamphetamine use.