Previous attempts to develop batterer risk assessment measures have produced weak prediction results. These approaches have generally relied on a dichomotous measure for reassault, specifically “reassault” or “no reassault,” and have eschewed the use of conditional or situational factors such as batterer or victim employment. The current study drew on a multisite longitudinal dataset of batterers and their female partners to develop and test a risk assessment instrument with multiple outcomes and conditional factors. The dataset included 840 men who were admitted to batterer programs in 4 cities; interviews were conducted with batterers and their female partners at program intake and every 3 months for a 15-month period. Measures included abuse and reassault outcomes, social background and relationship variables, batterer behavior and personality, victim characteristics, program characteristics, and conditional variables. Results of multinomial logistic regression analyses suggested that the use of multiple outcomes improved prediction results. However, conditional variables did not significantly improve the prediction model and the low “correct classification” indicated the difficulty of predicting the most dangerous batterers. The findings suggest the need for researchers and practitioners to distinguish between one-time reassault and repeat reassault. Future research should continue to search for additional risk markers to improve the prediction of reassault among batterers. Tables, references