The presence of occult axillary nodal metastases was evaluated in 159 patients with "node-negative" invasive breast carcinoma. Multiple additional levels of the lymph nodes were examined with hematoxylin-eosin staining and keratin immunostaining. Occult nodal metastases were detected in 50 (31%) patients; of these, 28 (17%) were detectable by hematoxylin-eosin stain alone, while the other 22 (14%) consisted of mostly single cells or very small clusters and required immunostaining for detection. The size of the metastatic deposit was < or = 0.2 mn in 31 (19%) patients and greater than 0.2 mm in 19 (12%) patients. Occult nodal metastasis correlated with the presence of peritumoral lymphatic invasion (P = .02) and was seen more frequently with larger tumor size, increased microvasculature, and aneuploidy. As a group occult metastases had no significant prognostic impact. However, patients with metastases measuring greater than 0.2 mm had significantly worse recurrence (P = .02), disease-free survival (P = .04), and overall survival (P = .07) rates; those with metastases detectable by hematoxylin-eosin stain alone also had a less favorable, although not significant, outcome. In contrast, patients with occult metastases that were < or = 0.2 mm or that were detected only by immunostaining had a survival rate comparable to and in fact slightly higher than that of the group without occult metastasis; 23 of these patients were without recurrence after a median follow-up of 11 years. Extension into perinodal soft tissue was an unfavorable feature. In a multivariate analysis peritumoral lymphovascular invasion and increased microvasculature were the most important prognostic parameters, and the presence of occult metastases greater than 0.2 mm was no longer significant. Our data suggest that occult metastases < or = 0.2 mm, especially those consisting of single cells, do not add useful prognostic information, and immunohistochemical studies to detect them are probably unnecessary. Larger metastases and extranodal involvement may have important prognostic value, but in this study they accounted for only 20% of patients who had recurrences or 6% of the total population. This underscores the importance of using more than one prognostic parameter in evaluating breast carcinoma.