The authors examine predictors of family participation in the G.R.E.A.T. Families Program of the Multisite Violence Prevention Project (MVPP), a four-site collaboration examining student, teacher, and family interventions for middle school students. Teachers recruited two cohorts of sixth grade students, recognized as being aggressive and influential with their peers, and their families into a voluntary, 15-session weekly program. Among the 643 families that consented (66%), the mean number of sessions attended was 8.13, with almost half (48.3%) attending 11 or more sessions. Linear mixed models (LMMs) were developed to predict the number of sessions attended based on a cluster of demographic variables and scales designed to measure aggression, problem behaviors, family factors, and other psychosocial constructs. Three of the nine clusters held significance when associated with attendance: level of aggression as rated by the child (negative association), parent-child bond (negative association), and level of child victimization (positive association). Somewhat surprisingly, the variance component due to interventionist turned out to be small and to constitute a nonsignificant component of the overall variability in attendance. Results suggest that family recruitment for multiple family group programs can be achieved with substantial effort and that resources available for recruitment might be differentially applied across families based on predictors of attendance. (Contains 4 tables and 1 footnote.)