This study demonstrates the utility of classification and regression trees (C&RTs) in predicting treatment outcome. A total of 177 HIV-positive survivors of childhood sexual abuse were randomly assigned to either a 15-session coping group condition or a 15-session support group condition. Six predictive categories were identified using C&RT, three predicting clinically significant improvement and three predicting no change. Variables predicting outcome differed by intervention type; depressive symptoms and active coping strategies predicted outcome for the coping group condition, and shame about sexual abuse and dissociative symptoms predicted outcome for the support group condition. Methodological issues and implications for clinical decision making are discussed.