The purpose of this study was to compare longitudinal smoking prevention program effects estimated on a population-based cohort sample of sixth- and seventh-grade students (average n per year = 4,664) using different schools as units of analysis (middle/junior high school as the school of origin or high school as the endpoint school of intervention). Fifty schools in 15 school districts were demographically matched and assigned to either a school and community-based program for prevention of cigarette, alcohol, and marijuana use, or a health education as usual control group. Smoking was measured by questionnaires administered to the students. An expired air (CO) measure of smoking was also administered to increase accuracy of self-reports. Program effects were estimated with regression analyses, controlling for school-level socioeconomic status, racial/ethnic make-up, urbanicity, and grade. Using school of origin as the unit of analysis, program effects showed 1 year net reductions of -8, -6, and -5% in prevalence rates of smoking in the last month, last week, and last 24 hr; 2-year program effects showed similar net reductions of -6, -5, and -3% (P's less than 0.10-0.001). Analyses with endpoint school as the unit showed slightly weaker effects at the 2-year follow-up, the year during which 64% of students had moved to a junior high or high school. The findings are discussed in terms of the potential contamination of experimental groups in longitudinal studies from school consolidations, family mobility, and feeder patterns to high schools, and the different smoking environments represented by middle, junior high, and high schools.