Objective: An exploratory analysis of self-report data gathered by the Family Connections program is used to build a predictive model of program completion. Method: The sample includes 136 families in a poor, urban neighborhood who meet risk criteria for child neglect. Families are randomly assigned to receive 3- or 9-month interventions. Bivariate analyses compare families who did and did not complete services. Logistic regression analysis identifies service-completion predictors for 136 families for whom pre- and post-data are available. Results: More caregivers in the 3-month group complete services. Completers have more children and report a more positive alliance with their workers. Depressive symptoms, worker alliance, and treatment-group status predict service completion in the final model. Conclusions: Findings and implications for practice are discussed. (Contains 2 tables.)