A total of 154 patients admitted for the first time to hospital for a psychotic or affective disorder in 1925 were rediagnosed in accordance with DSM-III and Leonhard's diagnostic system. Symptoms were rated in accordance with a rating protocol containing 33 items based on Leonhard's descriptions of cycloid psychoses. The patients were followed up through parish registers and hospital records. Of 42 cases considered to be cycloid psychosis at index admission, 34 were prognostically verified. The symptom ratings of the 154 patients were analysed by K-means cluster analysis to test whether the patients with cycloid psychoses would separate from the rest of the material. With a 2-cluster solution, 30 of 34 cases (88%) were contained in the same cluster. The sensitivity of the ratings was therefore judged to be acceptable. Specificity was low, however, since 19 cases in the cluster were differently diagnosed. A subcohort of 64 patients, satisfying at least 5 items of the rating protocol, was then analysed by Q-factor analysis to test whether nuclear cases of cycloid psychosis differ from symptomatically related syndromes. No such differences could be statistically verified; no symptom profile specifically indicating cycloid psychosis could be found. Prominent confusion symptoms appear, however, to be prognostically favourable, whereas motility symptoms without confusion seem to indicate an unfavourable course.