The aim of many research investigations is to compare the proportion of individuals in each of several groups that have a certain characteristic. The unit of allocation for such investigations is often an intact social unit, as in randomizing families, medical practices, schools, or entire communities, to different intervention groups. Standard statistical methods are not appropriate for these designs, since they do not take into account the dependencies among individuals within the same cluster. The authors review the strengths and weaknesses of several approaches for dealing with this problem, using data from a school-based smoking cessation trial. A principal conclusion is that the choice of method should depend on whether or not random allocation is used in the assignment of interventions.