Article dans une revue: Many years of concerted policy effort in Western countries has not prevented young people from experimenting with cigarettes, alcohol and marijuana. One potential explanation is that social interactions make consumption "sticky". We use detailed panel data from the Add Health survey to examine risky behavior (the consumption of tobacco, alcohol and marijuana) by American adolescents. We find that, even controlling for school fixed effects, these behaviors are correlated with lagged peer group behavior. Peer group effects are strongest for alcohol use, and young males are more influential than young females. Last, we present some evidence of non-linearities in social interactions.