Background. Increasing evidence suggests psychosis may be more meaningfully viewed in dimensional terms rather than as discrete categorical states and that specific symptom clusters may be identified. If so, particular risk factors and premorbid factors may predict these symptom clusters. Aims. (i) To explore, using principal component analysis, whether specific factors for psychotic symptoms can be isolated. (ii) To establish the predictors of the different symptom factors using multiple regression techniques. Method. One hundred and eighty-nine inpatients with psychotic illness were recruited and information on family history, premorbid factors and current symptoms obtained from them and their mothers. Results. Seven distinct symptom components were identified. Regression analysis failed to identify any developmental predictors of depression or mania. Delusions/hallucinations were predicted by a family history of schizophrenia and by poor school functioning in spite of normal premorbid IQ (F = 6.5; P <0.001); negative symptoms by early onset of illness, developmental delay and a family history of psychosis (F = 4.1; P = 0.04). Interestingly disorganisation was predicted by the combination of family history of bipolar disorder and low premorbid IQ (F = 4.9; P = 0.003), and paranoia by obstetric complications (OCs) and poor school functioning (F = 4.2; P = 0.01). Conclusion. Delusions and hallucinations, negative symptoms and paranoia all appeared to have a developmental origin though they were associated with different childhood problems. On the other hand, neither mania nor depression was associated with childhood dysfunction. Our most striking finding was that disorganisation appeared to arise when a familial predisposition to mania was compounded by low premorbid IQ. © 2002 Éditions scientifiques et médicales Elsevier SAS.