- Background. A prospective, 3-year cohort study of smoking habits was conducted among 9th grade pupils in Israel. Methods. A self-administered questionnaire was answered by pupils in the 9th grade and again in the 11th and 12th grades. Results. The baseline questionnaire was answered by 748 9th graders. In the 11th grade 448 (60%) completed the questionnaire for the second time, and in the 12th grade 388 (52%) completed the questionnaire. In all, 312 pupils (42% of the original cohort) completed all three questionnaires. There were no significant sociodemographic differences between these 312 and the original 748. Eight 9th graders (2.6%) were active smokers (at least one cigarette daily for the last month) compared with 64 (20.5%) in the 11th grade (P < 0.0001 vs 9th grade) and 70 (22.4%) in the 12th grade (P = 0.34 vs 11th grade). Most pupils knew the health hazards of smoking. Variables associated with smoking in the 11th grade included religiosity (religion was associated with lower smoking rates, P = 0.07), past experimentation with smoking (P < 0.0001), smoking among family members (P < 0.01), perceived future smoking status (P < 0.001), self-image (P < 0.001), influence of a teacher (P = 0.07) or celebrity (P < 0.05), and effect of peer pressure (P < 0.01). These results were similar for the 12th grade students except for active or previous smoking by gender (a significantly greater proportion of females than males, P = 0.04). Conclusions. The study design reduces the potential bias of follow-up attrition on assessment of predictors for smoking initiation. Gender, social modeling and peer pressure, past experimentation with smoking, smoking among family members, role models, and selfimage were associated with smoking. These factors should be emphasized in intervention programs.