The presenting clinical features of 163 previously untreated patients with multiple myeloma were correlated with survival duration using univariate and multivariate regression analyses. The univariate proportional hazard analysis ranked the parameters in the following order of importance: platelet count, haemoglobin level (Hb), tumour cell mass stage, lytic bone lesions, creatinine and age. When the individual contribution of each variable was assessed by multivariate regression analysis, platelet count was confirmed to be the dominant feature for prognosis, while clinical stage provided additional information. The introduction of platelet count could then be used to improve the discriminating power of Durie & Salmon staging, by allowing separation of the high-risk group (stages II and III) into a smaller subgroup (22%) of thrombocytopenic patients (less than 150 x 10(9) platelets/l) whose risk of death was actually very high (median survival: 9 months) and a larger subgroup (46%) of patients with normal platelet count and intermediate or standard risk (median survival: 48 months).