The assessment of efficacy of treatment in patients with recurrent glioma is notoriously difficult, and survival is the most objective endpoint. Between 1970 and 1992, a cohort of 211 patients with recurrent glioma received nitrosourea-based chemotherapy at the time of disease progression. The median survival from the start of chemotherapy was 7 months, with 30% 1-year and 10% 2-year survival probabilities. One-year survival was 22% in 147 patients with recurrent high-grade astrocytoma, 41% in 37 patients with low-grade astrocytoma and 45% in 24 patients with oligodendroglioma. Age, histological grade and Karnofsky performance status (KPS) at recurrence were independent prognostic factors for survival on multivariate analysis. Based on patients' age, tumour grade and KPS, it was possible to define three distinct prognostic groups with 1-year survival probabilities of 60, 21 and 17% (P < 0.005). Response to chemotherapy was difficult to assess but correlated with prognostic subgroup, with highest response rate (46%) in the most favourable group and lowest (13%) in the poor prognostic group. In patients with recurrent glioma, patient and tumour parameters are the major determinants of outcome which are identical to prognostic factors at the time of primary diagnosis. They can be used to provide prognostic information for the individual patient, and to stratify patients particularly in trials assessing the efficacy of novel treatments.