Evidence-based practice within the criminal justice system has relied primarily on cumulative reports of reassaults or rearrests, especially in evaluating effectiveness of domestic violence (DV) interventions. We use a longitudinal sample of DV offenders arrested and referred to DV offender programs in four cities to examine complex multi-outcome patterns of violent and abusive behavior and trajectories between patterns over time. Offenders and their partners were surveyed at program intake and followed for 15 months (N = 550). A Hidden Markov Model identifies four distinct behavioral patterns. Trajectory analysis indicates that the most abusive state is relatively stable and indicative of a high probability of future physical assault. Results suggest that evaluation based on complex outcomes may improve criminal justice intervention effectiveness, risk assessment, and risk management.