There is a strong empirical connection between individual and peer substance use during adolescence. The determination of whether this level of covariation reflects influence or selection is obscured by both the design and measurement strategies used. This present study utilizes a short-term longitudinal design with bi-monthly assessments to address the following two hypotheses: a) Adolescents select friends on the basis of their substance use, and b) New friend substance use predicts changes in future use. French Canadian adolescents (n = 143) were interviewed on their friendship networks and substance use behaviors (e.g., tobacco, alcohol and marijuana) four times during a school year. Cross-lag panel models revealed that adolescents who use substances tend to select new friends who use. Moreover, once in the network, these new friends also contribute to changes in the adolescents' substance use. These findings are relevant to understanding the multiple functions of adolescent substance use. (Contains 1 table and 1 figure.)