The Children's Cancer Study Group conducted prospective clinical trials of 608 children with non-Hodgkin's lymphoma from 1977 to 1983. In 1980, significant differences in survival of children with disseminated disease correlated with histologic diagnosis and the randomized treatment employed. A pathology reproducibility review showed the lymphoblastic lymphoma cases to be virtually 100 per cent distinguishable histologically from the nonlymphoblastic lymphomas (Burkitt's, non-Burkitt's, and "histiocytic"). However, diagnostic reproducibility of the pathologist-of-record was 59 per cent in the Burkitt's and non-Burkitt's lymphoma group. Therefore, 159 cases, agreed on by the pathologist-of-record and the "lymphoma panel" as Burkitt's (77 cases) or non-Burkitt's lymphoma (82 cases) and designated as the "reference diagnosis," were blindly reviewed twice each by two hematopathologists to yield the "review diagnoses." Consensus agreement was achieved in 67 per cent of cases overall, 82 per cent of Burkitt's and 54 per cent of non-Burkitt's lymphoma. Using the "reference diagnoses," we found that the relative frequency of Burkitt's and non-Burkitt's lymphoma was associated with the extent of disease at diagnosis (P = 0.06) but not with other prognostic factors. Despite the difficulties in histologic classification, analyses that used either "reference diagnoses" or "consensus review diagnoses" and that were adjusted for extent of disease consistently demonstrated significantly shorter event-free survival for patients having Burkitt's lymphoma; their failure rate was four times that for patient's with non-Burkitt's lymphoma. Newer cell biologic techniques hopefully will enhance histopathologic distinctions that remain the basis for diagnosis.