The ability to identify groups of children at risk of initiating a smoking habit may prove useful in developing effective smoking prevention programs. This report includes data collected over a three-year period, and attempts to predict adolescents' smoking behavior using both cross-sectional and longitudinal analyses. In addition, predictor variables reflecting both interpersonal and intrapersonal domains were included. Results indicated the highest rates of accurate classification into smoking categories were achieved with cross-sectional analyses. In addition, interpersonal variables emerged as most important in all analyses. Implications for smoking prevention programming are discussed.