Two hundred and fifty previously untreated adult patients with diffuse large-cell lymphomas were treated with a chemotherapy combination of cyclophosphamide, adriamycin, vincristine, prednisone, and low-dose bleomycin (CHOP-Bleo) with or without radiotherapy between 1974 and 1984. The 10-year survival rates for patients with Ann Arbor stages II, III, or IV disease of 55%, 42%, and 40%, respectively, were not significantly different. However, the survival rate of 76% for patients with stage I disease was clearly better. Factors more indicative of prognosis than stage, as found by univariant analysis, were tumor burden, serum lactic dehydrogenase level (LDH), age, and constitutional symptoms. From these, a multivariant analysis selected tumor burden, LDH level, and age as major independent factors for predicting survival (P less than .001). A prognostic risk model constructed on the basis of tumor burden and LDH levels identified four distinct risk groups (A, B, C, D) with 10-year survival rates of 85%, 66%, and 43% for A, B, and C. No patient in group D survived 10 years. These risk groups also had a strong correlation with complete remission rates and with relapse rates. Thus this model proved more effective for identifying patient populations according to their expected responses, durations of remission, and survivals than the Ann Arbor staging system. Detailed information supporting the use of this system for predicting prognosis and for treatment selection for patients with diffuse large-cell lymphomas is provided.