The objectives of the present study were to investigate the prognostic value of the International Prognostic Index (IPI) at relapse in the 215 patients with intermediate- or high-grade non-Hodgkin's lymphoma (NHL) included in the PARMA trial. The IPI at relapse was available in 204 (95%) of these patients. Response rates to 2 courses of DHAP were 77%, 54%, 55%, and 42% in patients with an IPI of 0, 1, 2 and 3, respectively (P <.02), whereas complete response (CR) rates were 33%, 29%, 20%, and 0% in the same groups of patients (P <.03). With a median follow-up period of 79 months, overall survival (OS) at 5 years was 46%, 25%, 25%, and 11% in these four groups (P <.001). One hundred nine patients responding to 2 courses of DHAP were randomized to receive either BEAC (carmustine, etoposide, cytarabine, cyclophosphamide and mesna) followed by autologous bone marrow transplantation (ABMT) or 4 additional courses of DHAP: IPI at relapse was found highly correlated to OS in patients treated in the DHAP arm (5-year OS: 48%, 21%, 33%, and 0% for IPI 0, 1, 2, and 3, respectively; P =.006), but not in the BEAC arm (5-year OS: 51%, 47%, 50%, and 50% for IPI 0, 1, 2, and 3, respectively; P =.90). OS was significantly superior in the BEAC arm as compared with the DHAP arm in patients with an IPI >0 (P <.05), but not in patients with an IPI of 0. In conclusion, these results show that IPI correlates to response and overall survival in patients with aggressive NHL in relapse and enables us to identify patients with a significantly different outcome among those treated with conventional chemotherapy alone.