Four prognostic models described for aggressive malignant lymphomas and the classical Ann Arbor staging system were used to compare the survival of 737 patients treated with the LNH-84 regimen. The aim of the study was to determine the optimal prognostic system at the time of diagnosis. Three institutions have described these models after multivariate analyses: the Dana Farber Cancer Institute (DFCI1 and DFCI2), the MD Anderson Hospital (MDAH), and the Memorial Sloan- Kettering Cancer Center (MSKCC). The models were constructed with the following variables: performance status, LDH level, and tumor extension. The latter is the most difficult to assess: it was considered as the number of extranodal sites and the diameter of the largest mass in DFCI1, stage and the diameter of the largest mass in DFCI2, the number of extranodal and extensive nodal sites in MDAH, and the number of nodal sites and their localization in MSKCC. Univariate studies with LNH-84 regimen patients showed all these variables to have major prognostic significance (logrank tests: P less than 10(-4)). All five prognostic systems divided patients into three subgroups: good, intermediate, and poor prognosis. Logrank analyses of survival showed highly significant differences (X2 greater than 90 and P less than 10(- 6)) between the subgroups. No gross difference was found between the models, and none was better than the others. A new, internationally accepted prognostic system for the expression and comparison of treatment results in aggressive malignant lymphomas should include major univariate prognostic parameters and must be reliable and easy to use in clinical practice. Until such time, stage or LDH level are the best alternatives.