The aims of this study were to determine if inter-school variation in smoking prevalence was due to differences in pupil composition or other school-level factors. A cohort of 13-14-year-olds (n = 7147) from 52 schools was followed-up 1 year later. Random effects logistic regression was used to examine school variation in smoking uptake and cessation, with and without adjustment for pupil composition. Inter-school variation in smoking prevalence is not caused by differences in pupil composition but is due to differences in the onset of smoking arising because of unmeasured school contextual or collective factors operating on pupils' decisions.