One reason for the inconsistent findings in schizophrenia research is the lack of diagnostic conformity. In the face of modern operational "atheoretical" diagnostic systems, this dilemma is still present. In order to examine specificity and validity of diagnoses, we carried out a systematic twin study with index twins suffering from schizophrenic spectrum psychoses. We compared the diagnostic systems of DSM-III-R, which is based on consensus of international experts, with Leonhards' nosology developed on sophisticated clinical observation and description of psychopathological phenomena occurring during the long-term course of psychiatric diseases. We examined twin concordance, family history, and the frequency and severity of complications of pregnancy and child-birth. The results suggest that the schizophrenic spectrum has to be divided into clinically and etiologically heterogeneous subgroups. This was much more striking when Leonhard's diagnostic criteria were used than with DSM-III-R diagnostic criteria. There seem to be three valid and etiologically different groups: cycloid psychoses, unsystematic schizophrenias and systematic schizophrenias as proposed by Leonhard. In cycloid psychoses genetic loading seems to be low (proband concordance MZ 38%, DZ 29%), but pregnancy and birth complications may have an important role in the etiology. On the other hand, unsystematic schizophrenias are obviously predominantly inherited (proband concordance MZ 88%, DZ 17%) and "environmental" factors are not very prominent. It is striking that MZ twins with a diagnosis of systematic schizophrenia have not yet been found, whereas 32% of DZ index twins (6 out of 19) were diagnosed as having systematic schizophrenia. Further, all DZ twins with the diagnosis of systematic schizophrenia were discordant and the affected twins had threetimes as many and as severe pregnancy and birth complications in the history than their healthy co-twins.