We assessed predictors of self-reported excessive drinking (> 5 drinks) in a sample of heavy drinkers. Participants were randomly assigned to moderation training or a waiting-list control condition. They were trained in ecological momentary assessment (EMA) involving self-monitoring of drinking and other variables on a small hand-held computer, the electronic diary (ED). During the 8-week study, participants were compliant in their use of the ED for both random prompts and the entry of data related to specific drinking episodes. Generalized estimating equations were used to fit models involving predictors related to past history of drinking, aspects of the training program, drinking restraint, and episode-specific mood. The models indicated robust predictors of decreased and increased drinking. Our results suggest that EMA is a useful methodology for assessing drinking and related behaviors.