Two hundred and ninety-seven seventh grade students (143 males and 154 females) participated in a prospective study to predict adolescent cigarette smoking behavior one year later. Predictor variables included 10 survey items assessing the smoking behavior of students' friends and family, students' school behavior and beliefs about smoking and students' intentions to smoke in the future. The one-year follow-up survey was administered under "bogus pipeline" conditions to enhance the validity of self-reported smoking status by including the collection of saliva thiocyanate samples. Univariate analyses indicated that smokers differed from nonsmokers on a number of measures and that there were few sex differences on either the survey variables or on smoking status. Stepwise discriminant function analyses revealed that it was possible to accurately predict the onset of adolescent cigarette smoking by combining the survey variables. The three variables that consistently accounted for the greatest proportion of te variance were number of friends who smoke, intentions to smoke in the future and percent of older siblings who smoke. The implications of this work for programs intended to prevent adolescent smoking are discussed.