Seven disjoint clusters were obtained from 340 MMPI's by an innovative cluster strategy that combined Ward's hierarchical clustering with a partitioning method. Cluster groups differed on racial composition and diagnoses. It was determined that the lack of differences among cluster groups on other clinically relevant variables may be due to the choice of measures and the sample's homogeneous nature. Demographic variables showed considerable similarities between the sample group and prison populations. The analysis further suggested that factors such as sociopathy, substance abuse, psychosis with paranoid features, and a history of criminal activities distinguished these offenders from the benign mentally ill. In general, the lack of differences between cluster groups on most variables indicated that the sample population cannot be subclassified meaningfully on the basis of MMPI configurations alone. It would probably be more profitable for investigators to use more direct measures of the variables of interest. 9 references and 2 tables