In the National Surgical Adjuvant Study for Breast Cancer (NSAS-BC), node-negative breast cancers were divided into higher- and lower-risk groups according to the histopathological nuclear grade given at individual collaborating hospitals, and the higher-risk group was entered into a randomized protocol of adjuvant therapy. Because the nuclear grade was the composite of nuclear atypia and mitotic counts, maintenance of interobserver agreement in mitotic counts was indispensable for the success of the protocol study. Fourteen pathologists participating in the protocol judged whether or not 20 photomicrographs suspected of showing mitotic cancer-cell figures truly showed mitoses. After standardizing the counting method, these pathologists counted the number of mitotic figures per 10 high-power fields of hematoxylin-eosin-stained main-tissue sections of 20 tumors. Areas where mitotic counts were considered to be the most frequent by each pathologist were compared for these tumors. For the judgment of whether the photomicrograph indicated mitosis, the level of interobserver agreement was moderate (kappa = 0.569). In the observations of 20 tumors, interobserver agreement level of mitotic counts was moderate (kappa = 0.506), that of nuclear atypia scoring was fair (kappa = 0.265), and that of nuclear grading was substantial (kappa = 0.633). The counted area was almost the same among the observers in 9 tumors, split into two areas in 6, and dispersed in 5. Concordance in judgment was achieved in 7 of the first 9 and in all of the third 5, but only in one of the second 6. The cause of discordance was mostly derived from tumor heterogeneity and the difference in the site where mitoses were counted. Interobserver agreement level was considered to be satisfactory, and it was expected that the case entry would be performed appropriately in the protocol study. The selection of the counting area was confirmed to be important for the acquisition of high-level agreement level in mitotic counts.