This study examined how perceptions of student smoking in the school environment and the actual smoking rate among senior students at a school are related to smoking onset. Multilevel logistic regression analysis was used to examine correlates of ever smoking in a sample of 4,286 grade 6 and 7 students from 57 elementary schools in Ontario, Canada. Students are at increased risk for smoking if they (a) often see students smoking near their school, (b) report that students at their school smoke where they are not allowed, and (c) attend a school with a relatively high senior student smoking rate. Each 1% increase in the smoking rate among grade 8 students increased the odds that a student in grades 6 or 7 was an ever smoker versus never smoker (odds ratio, 1.05; 95% confidence interval, 1.02-1.08). A low-risk student (no family or friends who smoke) was over twice as likely to try smoking if he/she attended a high-risk school. Prevention programs should target both at-risk schools and at-risk students, and strongly enforced policies preventing students from smoking on or near school property should be implemented.