Although the prognosis of high grade malignant glioma patients is generally poor, it is possible to identify groups of patients with varying prognoses. Basing our results on the first MRC glioma study, multivariate methods were used to identify prognostic factors independently associated with the length of survival. Young age, the presence of fits, especially of long duration, extensive surgical removal of tumour and good clinical performance status were found to be the most important predictors of longer survival. The effect of tumour grade (3 or 4) was not significant, being considerably diluted by an association with extent of neurosurgery. A prognostic index was derived which split the patients into 6 groups of varying prognoses, with 2-year survival rates of between 1 and 32%. The results were verified in patients entered into a subsequent MRC trial. The successful identification of different prognostic groups suggests the use of this index as an aid in making treatment decisions for individual patients, and in interpreting the results of uncontrolled phase II studies.