It has frequently been questioned whether results of trials can be generalized to routine clinical practice. Results obtained with standard cyclophosphamide, doxorubicin, vincristine, prednisone (CHOP) chemotherapy for aggressive non-Hodgkin's lymphoma in the control arm of prospective randomized phase III clinical trials during the past 25 years appear to be comparable. As the possibility to generalize trial results is favored by such consistency, we tested this hypothesis and tried to indicate explanatory 'moderator variables' (inclusion and exclusion criteria, therapy characteristics, sample sizes, inequalities in the distribution of patients over prognostic categories across different trials) in the case of divergent trial results. Trial results on conventional CHOP chemotherapy were obtained from literature research. Overall response (OR), complete response (CR) and two-year overall survival (2YS) were considered as outcome measures. Although OR rates and 2YS rates were within acceptable limits of comparability, the absolute differences within the results were remarkably wide, particularly with regard to CR and 2YS. Divergent rates could not be properly explained by differences in the moderator variables. We conclude that absolute results obtained with CHOP in the control arm of trials appeared to be poorly generalizable to routine clinical practice, particularly in the case of elderly patients. This analysis underlines the need for the strict application of internationally agreed response criteria and the WHO classification system, large sample sizes and stratifying patients on the basis of prognostic factors, preferably in intergroup clinical trials. We expect those factors to lead to a better consistency of results in future trials and improved possibilities to generalize the results.