In a retrospective review of 76 consecutive cases of female breast cancer with no distant metastases at the time of the original biopsy, we compared how efficiently three different counting methods--mitoses/10,000 cells, mitoses/mm2 of tumor, and mitoses/10 high power fields (x 400)--would discriminate survivors from nonsurvivors 10 years after the original diagnosis. The effect was compared in subgroups of patients with different axillary lymph node status. The conclusions are that: (a) the methodology for counting mitoses is not trivial. Indeed, only mitoses/mm2 of tumor permitted prognostic predictions approaching overall statistical significance, at the exclusion of mitoses/10,000 cells, and mitoses/10 high power fields. (b) In each subgroup based on lymph node status, a low mitotic index in the primary tumor confers a better prognosis than a high mitotic index. (c) When comparing subgroups defined by increasing number of metastatic lymph nodes, one observes a gradually increasing proportion of patients whose primary tumor displays a high mitotic index, thus suggesting that a high mitotic index in the primary tumor predisposes to metastases. (d) A combination of lymph node status and mitotic counts offer a more refined ranking for risk of tumor recurrence than predictions based on lymph node status alone. (e) Although mitotic activity has a measurable impact on disease outcome of a patient population, it is paradoxical that counting mitoses is of disappointing prognostic value as a clinical tool. This is so because the method is not helpful in predicting outcome in those individuals whose disease will evolve contrary to the expectations conferred by their lymph node status.