This study reports 3-year outcomes for clients who had been treated in the five outpatient sites of Project MATCH, a multisite clinical trial designed to test a priori client treatment matching hypotheses. The main purpose of this study was to characterize the status of the matching hypotheses at the 3-year follow-up. This entailed investigating which matching findings were sustained or even strengthened across the 3-year study period, and whether any hypotheses that were not supported earlier eventually emerged at 3 years, or conversely, whether matching findings discerned earlier dissipated at this later time. This research also examines the prognostic effects of the client matching attributes, characterizes the overall outcomes at 37 to 39 months, and explores differential effects of the three treatments at extended follow-up. With regard to the matching effects, client anger demonstrated the most consistent interaction in the trial, with significant matching effects evident at both the 1-year and 3-year follow-ups. As predicted, clients high in anger fared better in Motivational Enhancement Therapy (MET) than in the other two MATCH treatments: Cognitive-Behavioral Therapy (CBT) and Twelve-Step Facilitation (TSF). Among subjects in the highest third of the anger variable, clients treated in MET had on average 76.4% abstinent days, whereas their counterparts in the other two treatments (CBT and TSF) had on average 66% abstinent days. Conversely, clients low in anger performed better after treatment in CBT and TSF than in MET. Significant matching effects for the support for drinking variable emerged in the 3-year outcome analysis, such that clients whose social networks were more supportive of drinking derived greater benefit from TSF treatment than from MET. Among subjects in the highest third of the support for drinking variable, TSF participants were abstinent 16.1% more days than MET participants. At the lower end of this variable, difference in percent days abstinent between MET and TSF was 3%, with MET clients having more abstinent days. A significant matching effect for psychiatric severity that appeared in the first year posttreatment was not observed after 3 years. Of the 21 client attributes used in testing the matching hypotheses, 11 had prognostic value at 3 years. Among these, readiness-to-change and self-efficacy emerged as the strongest predictors of long-term drinking outcome. With regard to the overall outcomes, the reductions in drinking that were observed in the first year after treatment were sustained over the 3-year follow-up period: almost 30% of the subjects were totally abstinent in months 37 to 39, whereas those who did report drinking nevertheless remained abstinent an average of two-thirds of the time. As in the 1-year follow-up, there were few differences among the three treatments, although TSF continued to show a possible slight advantage.