Smoking rates in adolescents are rising. Helping young people to avoid starting smoking is a widely endorsed goal of public health, but there is uncertainty about how to do this. Schools provide a route for communicating with a large proportion of young people, and school-based programmes for smoking prevention have been widely developed and evaluated. To review all randomised controlled trials of behavioural interventions in schools to prevent children (aged 5 to12) and adolescents (aged 13 to18) starting smoking. We searched The Cochrane Controlled Trials and Tobacco Review group registers, MEDLINE, EMBASE, Psyclnfo, ERIC, CINAHL, Health Star, Dissertation Abstracts and studies identified in the bibliographies of articles. Individual MEDLINE searches were made for 133 authors who had undertaken randomised controlled trials in this area. Types of studies: those in which individual students, classes, schools, or school districts were randomised to the intervention or control groups and followed for at least six months. Children (aged 5 to12) or adolescents (aged 13 to18) in school settings. Types of interventions: Classroom programmes or curricula, including those with associated family and community interventions, intended to deter use of tobacco. We included programmes or curricula that provided information, those that used social influences approaches, those that taught generic social competence, and those that included interventions beyond the school into the community. We included programmes with a drug or alcohol focus if outcomes for tobacco use were reported. Types of outcome measures: Prevalence of non-smoking at follow-up among those not smoking at baseline. We did not require biochemical validation of self-reported tobacco use for study inclusion. We assessed whether identified citations were randomised controlled trials. We assessed the quality of design and execution, and abstracted outcome data. Because of the marked heterogeneity of design and outcomes, we did not perform a meta-analysis. We synthesised the data using narrative systematic review. We grouped studies by intervention method (information; social competence; social influences; combined social influences/social competence and multi-modal programmes). Within each category, we placed them into three groups according to validity using quality criteria for reported study design. Of the 76 randomised controlled trials identified, we classified 16 as category one (most valid). There were no category one studies of information giving alone. There were fifteen category one studies of social influences interventions. Of these, eight showed some positive effect of intervention on smoking prevalence, and seven failed to detect an effect on smoking prevalence. The largest and most rigorous study, the Hutchinson Smoking Prevention Project, found no long-term effect of an intensive 8-year programme on smoking behaviour. There was a lack of high quality evidence about the effectiveness of combinations of social influences and social competence approaches. There was limited evidence about the effectiveness of multi-modal approaches including community initiatives. There is no rigorous test of the effects of information giving about smoking. There are well-conducted randomised controlled trials to test the effects of social influences interventions: in half of the group of best quality studies those in the intervention group smoke less than those in the control, but many studies showed no effect of the intervention. There is a lack of high-quality evidence about the effectiveness of combinations of social influences and social competence interventions, and of multi-modal programmes that include community interventions.