Aims. To examine the relationship between various interpersonal, intrapersonal and situational factors assessed at 6, 12 and 18 months after entrance to continuing care, and cocaine use in subsequent periods. Design. A randomized clinical trial with a 2-year follow-up. Setting. An outpatient US Veterans Administration substance abuse treatment program. Participants. Cocaine-dependent male veterans (N = 132) entering continuing care. Interventions. 12-Step focused group treatment vs. individualized relapse prevention continuing care. Measurements. Motivation, coping and mood, social support, co-morbid problem severity, treatment attendance, self-help participation and cocaine use variables were assessed at each follow-up. Findings. During the 2-year follow-up, patients used cocaine on fewer than 8% of the days in each of the four 6-month periods, which represented a highly significant decrease in relation to cocaine use prior to treatment. In univariate analyses, abstinence commitment, self-efficacy, positive mood, support from family, employment, attendance in continuing care and self-help participation at two or more of the follow-ups predicted less cocaine use in subsequent 6-month periods. Readiness to change and medical, psychiatric, legal and family/social problem severities either did not predict, or were inconsistently related to, subsequent cocaine use. In multivariate analyses, degree of self-help participation emerged as the strongest and most consistent predictor of cocaine use. However, when current cocaine use at each follow-up was controlled, none of the predictors was significant at more than one follow-up point. There was little evidence of interactions between treatment condition and the predictor variables. Conclusions. Continued self-help participation and the early achievement of cocaine abstinence appear to be important factors in the maintenance of good cocaine use outcomes over extended periods. The results also highlight the importance of controlling for various post-treatment factors when evaluating the relationship between any one factor and subsequent outcome, as many of the factors that were significant predictors in the univariate analyses were no longer significant when other factors were controlled.