Despite agreement that chronic schizophrenia is clinically heterogeneous, no system of subclassification has received general acceptance. Most methods for defining subgroups have relied either on course or symptom type alone. Using cluster analysis on patients rated on a broad range of items derived from commonly employed explicit diagnostic criteria, we were able to define two reasonably distinct groups. The first and larger group was characterized by later onset, good premorbid adjustment, and well-organized delusions; the second by poor premorbid adjustment, early onset, family history of schizophrenia, and symptoms such as bizarre behavior, incoherent speech, and blunted effect. Two different clustering programs yielded similar profiles. Analysis of computer-simulated data indicated that such good agreement would occur by chance less than one time in a hundred. Furthermore, the groupings remained stable when extra cases were added to the sample, and affected family members showed a significant tendency to occur in the same cluster. Although an etiological distinction between the subtypes cannot be inferred, we suggest that this form of classification provides a potentially useful starting point for further biological and genetic research.