Complete axillary dissection was performed in 287 patients undergoing modified radical mastectomy between 1984 and 1987 to identify patterns of axillary node metastases, as well as discontinuous axillary node ("skip") metastases. Positive pathologic findings were compared with preoperative clinical examinations in 266 patients and showed only 60 cases (22.6%) clinically suspicious for tumor, in contrast to 131 (45.6%) with pathologically confirmed positive lymph nodes. Axillary contents were classified level I, II, or III based on their relationship to the pectoralis minor muscle. An average of 24.2 nodes was resected per patient (level I, 10; level II, 8.1; and level III, 5.3). Tumors ranged in size from 0.5 to 12.0 cm (mean, 2.6 cm), and increasing tumor size was associated with an increased likelihood of positive nodes. The data on 204 patients with complete clinical and pathologic data show that of 119 patients with negative level I nodes a limited axillary dissection (level I only) would fail to identify 6 with positive level II and 2 with positive level III nodes, whereas of 85 patients with positive level I nodes limited axillary dissection would fail to identify 17 with positive level II nodes, 7 with positive level III nodes, and 27 with positive levels II and III nodes. Complete axillary dissection (levels I, II, and III) should be performed to stage patients accurately, as well as to remove tumor-involved nodes and diminish local axillary recurrences. Clinical examination of the axilla appears to be a poor means of identifying axillary metastatic cancer.