Statistical modelling was used to analyse clinical features and the Ki-67 proliferation index in a study of 77 cases of glioblastoma multiforme. Relative youth, frontal tumour site and treatment with external beam radiation had an important positive influence on survival. The pre-operative Karnovsky score and the presence of necrosis were not related to outcome. Despite the attractive hypothesis that rapidly proliferating tumours might be associated with a worse prognosis compared with slowly proliferating lesions, the Ki-67 index in this study offered no prognostic information even when individual sites were considered separately. The use of this form of computer modelling and its role in analysis of prognostic data is discussed.