This study examined relations between group process variables and retention of ethnic minority (African American and Hispanic) caregivers in a family-focused preventive intervention. Data from the Familias Unidas/SEPI project (Coatsworth, Pantin, & Szapocznik, 2002), a randomized, controlled intervention trial, were used to cluster participants according to their patterns of retention over 30 intervention sessions. These person-centered analyses identified three broad patterns: (a) dropouts; (b) variable-attenders; and (c) consistent-high-attenders. Two subgroups of the variable-attender group were also identified: (a) intermittent-attenders, and (b) continual-attenders. Discriminant Function Analysis (DFA) with follow-up Analysis of Variance tested for differences among the three main retention groups on facilitator ratings of participants' general level of participation, leadership, positive alliance with the group, and negative alliance with the group during the first half of the intervention. Leadership and positive alliance significantly discriminated the broad retention patterns. Mean level of participation was not significantly different across retention groups. Results of DFA and ANOVA analyses using leadership, alliance, and participation variables from the first and second halves of the intervention indicated only leadership and positive alliance during the second half of the intervention discriminated continual-attenders from intermittent-attenders. EDITORS' STRATEGIC IMPLICATIONS: The authors describe a promising approach to studying facilitators' assessments of client involvement in a family-focused preventive intervention. The quality of the participants' behavior during sessions, rather than their absolute levels of participation, predicted their pattern of retention in the program. Future comparisons of facilitator and parent views may prove helpful.