Clinical data from two unselected series of patients were subjected to cluster analysis in order to find out whether clusters of patients corresponding to existing diagnostic categories could be obtained. Each series consisted of 250 consecutive admissions, one to an English area mental hospital and the other to an American state hospital. From the extensive clinical data available for each patient 70 items were chosen, and these were transformed into ten factors by principal component analysis. Both the American and the English data were then subjected to two different methods of cluster analysis—Wolfe's Normap and Everitt's Class 1—using scores on these ten components as input.

All four analyses produced separate clusters identifiable with the manic and depressive phases of manic-depressive illness, with acute paranoid schizophrenia and with chronic or residual schizophrenia. Of these the manic and paranoid schizophrenic were the most and the chronic schizophrenic the least well defined. Patients with depressive and other neuroses, with personality disorders, and with alcoholism showed no tendency to form distinct clusters in any of the four analyses.

The significance of these findings for psychiatric nosology is discussed. The finding of relatively well defined clusters of patients in this and other similar studies suggests that within the field of psychotic illness a typology may be the most appropriate form of classification, in spite of the lack of evidence for the existence of genuine boundaries between syndromes. Perhaps the most useful form of classification will eventually prove to be a combination of a dimensional system with a typology.