The objectives of this review were to examine whether smoking prevalence varies between schools independently of health promotion programmes and pupil composition, to show which school characteristics are responsible for this variation, and to examine the methodological adequacy of such studies. Searches for published studies were performed on medical, educational and social science databases, relevant articles' reference lists, and citation searches. Any study was included that described inter-school variation in smoking prevalence, or related such variation to school characteristics. A model relating pupil smoking to school, neighbourhood, and pupil characteristics unlikely and likely to be influenced by school was used to examine the adequacy of control of confounding by pupil composition. Data from studies were combined qualitatively considering methodological adequacy to examine the relation of smoking prevalence to school characteristics. Theoretical frameworks underpinning the choice of school characteristics and postulated relationships between these characteristics and smoking prevalence were described. There were large variations in smoking prevalence between ostensibly similar schools. Evidence that pupil composition did not cause this was weak, because all studies had methodological problems, including under control of relevant pupil compositional factors and over control of factors likely to represent the mechanism through which schools influence pupils' smoking. There was little evidence that elements of tobacco control policy other than bans and enforcement deterred smoking. Academic practice and school ethos were related to smoking. Academically selective schools did not influence smoking, once pupil composition was controlled. There was one study on neighbourhood influences, which were unrelated to smoking. Studies frequently offered little or no theoretical justification for associating school characteristics with smoking. Some aspects of school influence pupils' smoking, probably independently of pupil composition. However, under-control and over-control of confounding and lack of theoretical underpinning precludes definitive conclusions on how particular school characteristics influence pupils' smoking.