In search of features of prognostic importance, a cohort of patients admitted to a mental hospital in 1925 was investigated by means of multivariate clustering techniques. Using K-means cluster analysis or Q-factor analysis, a group containing cases with unfavourable prognosis was isolated. Other groups derived were prognostically heterogeneous. One group of patients, in early phases similar to good prognosis schizoaffective psychoses, could be distinguished and characterized by non-symptom items. There was initial periodicity and onset was acute. They were, on average, younger than the other subjects and there was no personality deviation or emotional disturbance before onset of disease. A family history of mental illness was rare. Two of the factors were positively and negatively characterized by items covering familial history of mental illness, thus seemingly confirming the familial vs sporadic distinction in the subclassification of schizophrenia. Though the clinical pictures were distinctively different at the time when the ratings underlying the analysis were made, approximately the same proportion of cases in the two groups had independently been diagnosed as paranoid schizophrenia--also taking the course of illness into account. It could furthermore be shown that the population at risk--siblings and children of subjects--as well as the observed number of years at risk in these groups were significantly smaller in the sporadic group than in the familial group. This was a combined effect of a lower fertility in subjects and parents in the sporadic group and a higher rate of drop out due to mortality and other reasons among siblings of these subjects. The same tendency was indicated when subjects with and without family history irrespective of factor belongingness were compared. It cannot be concluded that the familial vs sporadic distinction is without relevance in the research on schizophrenia, but its essence may easily be obscured, if the population at risk is not taken into account.