Risk factors for the uptake of cigarette smoking were examined prospectively in 2159 non-smoking secondary schoolchildren aged 11-13 who participated in a survey in 1983 and were followed up 30 months later, by which time 35 per cent had taken up smoking. In a multivariate logistic model, the strongest predictors to emerge were prior experimentation with cigarettes and sex, with more girls (41%) than boys (30%) starting to smoke. Other predictors of taking up smoking were being uncertain about smoking in the future, reporting having been drunk, having a boy or girl friend, believing teachers and friends would not mind if they took up smoking, and giving lower estimates of prevalence of smoking among teachers. Parental smoking behaviour and attitudes, beliefs about the effects of smoking on health, opinions about smoking and perceived strictness of parents did not predict take up of smoking when other variables were controlled for. The odds of taking up smoking varied from 0.24 (risk = 0.19) for a child with the most favourable combination of risk factors to 3.49 (risk = 0.78) for a child with the worst prognosis. These results differ from those of many cross sectional studies and hence indicate the importance of a prospective approach to this type of research.