OBJECTIVE To study the development of smoking behavior in adolescents using a longitudinal, multivariate design. METHODS Adolescents (n = 14,133, age range 12 to 18 years) took part in the longitudinal Add Health study (two waves, separated by 1 year, 56% smokers and 44% nonsmokers at Wave 1). Eight risk factor domains were established at Wave 1 (daily activities, psychological health, personality, school situation, family functioning, rough living, religion, and neighborhood status), which were further separated into subdomains by factor analysis. Subdomains were used to predict risk at Wave 2 of smoking initiation, progression, or failure to discontinue, using logistic regression analysis. Analyses were performed for boys and girls separately and results corrected for age, race, urbanicity, and socioeconomic status. RESULTS Use/abuse of other substances by self and peers influenced most stages of smoking, whereas trouble in school was associated with initiation and progression of smoking. Poor family relations predicted initiation of experimental smoking for girls, whereas low involvement in active pastimes predicted failure to discontinue experimental smoking. For boys, low religiosity predicted progression to regular smoking and failure to quit regular smoking, whereas delinquency also reduced success of regular smoking discontinuation. CONCLUSIONS These findings may direct efforts for prevention and intervention of adolescent smoking behavior and may also provide guidance for future studies.