An international, multi-institutional, clinical, pathologic study of 1175 cases of non-Hodgkin's lymphoma sponsored by the National Cancer Institute has offered a unique opportunity to investigate issues of reproducibility and agreement in pathologic classification. Teams of experienced hematopathologists reviewed all 1175 cases, with the "experts" utilizing their own classification system and the experienced panelists utilizing all six of the histopathologic classification systems. The fact that 20% of the slides were randomly repeated for a second interpretation allowed assessment of the reproducibility of any one given pathologist. In addition, the agreement between and among pathologists was evaluated. Results of this retrospective study showed that both experts and panelists were able to identify follicular (nodular) or other indolent lymphomas with 95% probability of concurrence between initial and later interpretations. Moreover, individual pathologists agreed with others in identifying and classifying these follicular features with approximately 90% probability of agreement. The reproducibility for individual pathologists, based on the 20% of cases that were randomly repeated, varied from a probability 0.53 to 0.93. Comparisons of intersystem predictability demonstrated that no one system predicted for any of the other systems within any major degree of reliability. This inability to "translate" from one system to another is thought to be a reflection of the problems of both interpathologist agreement and intrapathologist reproducibility. Flaws in study design are discussed and the usage of the working formulation of non-Hodgkin's lymphomas, in addition to classification by traditional schemas, is strongly encouraged.