This article presents the results of a meta-analysis designed to test the prevailing view that we largely understand why adolescents start to smoke and how to delay it. This view has developed even though none of the major reviews of the last 12 years has adjusted for the important methodological problems that all of those reviews identified as common in the published literature. School-based smoking prevention programs based on peer or social-type programs, published between 1974 and 1991, were included in this meta-analysis. Treatment characteristics were used to predict an effect size after adjustment for study design and population characteristics, and in particular, after a post hoc correction for errors in the original unit of analysis. The results suggest that the average effect for peer or social-type programs is likely to be quite limited in magnitude, and that the reduction in smoking may be only 0.10 standard deviation units, or perhaps 5%. Even under optimal conditions, the reduction in smoking may be only 0.50 to 0.75 standard deviation units, or perhaps 20%-30%.