Recent approaches to empirical derivation and validation of alcoholic subtypes which involve cluster analysis wihin a single multivariate assessment domain have been questioned. To investigate more clinically meaningful subtypes in a sample of 245 male alcoholics, cluster analysis was performed on a combination of variables from four assessment domains: demographics, drinking behavior, neuropsychological functioning, and psychopathology. Six subtypes were derived and shown to have discriminant validity with respect to measures which had not been used in the original cluster analysis. The six subtypes were found to have differential patterns of treatment outcome over 9-month follow-up, but only among patients randomly assigned to 2-week as opposed to 7-week hospitalization. Cluster membership was not associated with marital status or reason for referral, but was related to the number of previous treatments for alcohol-related problems or psychiatric reasons, drinking-related locus of control, and physical symptoms. The results suggest that empirically derived typologies based on multiple assessment domains should be investigated for their usefulness in more appropriately matching treatments to patient subtypes. (Author/JAC)