Most adolescent smoking prevention studies employ designs in which classrooms, schools, school districts, or sometimes whole communities are assigned to treatment conditions while observations are made on individual students. The critical design feature in such community trials is the nesting of intact social groups within treatment conditions. This combination requires that the treatment effect be assessed against the between-group variance; unfortunately, that variance is usually larger than for randomly constituted groups and its precision is usually less than that for the within-group variance. These factors often combine to reduce power so that it is almost impossible to detect important treatment effects in an otherwise well designed and properly executed study. To address these problems, investigators need good estimates of the intraclass correlation for the variables of interest, which together with the number of observations per unit determine the magnitude of the extra variation in the nested design. The purpose of this paper is to describe the methods and results from a study designed to generate estimates of intraclass correlation for common outcomes in adolescent smoking prevention studies and to discuss the use of these estimates in the planning of new studies.