An empirical clustering technique was applied to data obtained from 321 male and female alcoholics to identify homogeneous subtypes having discriminative and predictive validity. The clustering solution identified two "types" of alcoholics who differed consistently across 17 defining characteristics in the male and female samples. One group, designated type A alcoholics, is characterized by later onset, fewer childhood risk factors, less severe dependence, fewer alcohol-related problems, and less psychopathological dysfunction. The other group, termed type B alcoholics, is characterized by childhood risk factors, familial alcoholism, early onset of alcohol-related problems, greater severity of dependence, polydrug use, a more chronic treatment history (despite their younger age), greater psychopathological dysfunction, and more life stress. The two types also differed with respect to treatment outcome assessed prospectively at 12 and 36 months. The results are consistent with historical and contemporary typological theories that have postulated similar subgroups of alcoholics. The findings suggest that an empirically derived, multivariate typology of alcoholism has theoretical implications for explaining the heterogeneity among alcoholics and may provide a useful basis for predicting course and estimating treatment response.