This study prospectively followed 103 bipolar patients enrolled in a VA treatment program for 1 year. Mental health service utilization was recorded and aggregated using the VA Cost Distribution Report. We hypothesized that previously reported predictors of disease severity would also predict service utilization, and that several other predictors of service utilization might also be identified.Analyses indicated that only the presence of a major affective episode at clinic intake and a recalled history of childhood physical abuse predicted mental health service utilization. Contrary to expectations, previously reported predictors of disease severity were not significant predictors. Implications for the study of economic outcome are discussed. In particular, we propose that economic outcome is a complex function of both patient and system factors, rather than simply being passively driven by disease severity.