Relative operating characteristic (ROC) analysis was used to examine the clinical applicability of 3 breast carcinoma tumor markers, CA 15.3, carcinoembryonic antigen, and mucin-like carcinoma-associated antigen. Each tumor marker was quantitated in single serum samples collected from 100 normal blood donors, 60 patients with nonmalignant diseases, 33 women at high risk for breast carcinoma, 30 patients with malignancies other than breast carcinoma, and 158 breast carcinoma patients including 67 with no evidence of disease following surgery, 46 with a tumor burden less than 5 g, and 45 with a tumor burden greater than 5 g. These were used to construct models for early diagnosis among those at high risk for breast carcinoma, the influence of nonmalignant disease on early diagnosis, discrimination of breast carcinoma from other adenocarcinomas, detection of early recurrence, and assessment of change in tumor burden. For each model ROC data permitted the unbiased selection of the most appropriate critical values based on the interaction of sensitivity and specificity. ROC analysis indicated that in practice the assays were remarkably similar. While CA 15.3 generally performed best, there was significant variation among models. Optimal marker selection can thus depend on specific clinical application. In some cases ROC identified a combination of markers as superior to any single assay, but this was not statistically significant.