BACKGROUND Expression of carcinoembryonic antigen (CEA) has been reported in 10-95% of breast cancer. Its value as a predictor of disease progression is controversial. METHODS The expression of CEA in 202 Stages I and II breast carcinomas was assessed by immunohistochemistry, and the results were correlated with various histologic and clinical parameters to establish CEA's biologic relevance. The mean follow-up of the patients was 6.5 years. The monoclonal antibody used does not cross-react with other molecules in the CEA gene family. RESULTS One hundred, thirteen (56%) tumors expressed CEA in more than 15% of the cells. Expression of CEA was associated with positive estrogen receptor (ER) status (P = 0.003). Univariate Cox regression analysis showed that, whereas disease free survival (DFS) and overall survival (OS) were not associated significantly with CEA expression, tumor size, nuclear grade, ER status, lymph node metastases, and stage were. When ER status was stratified to CEA expression, patients who were ER negative and had CEA-negative tumors had a 3.9 times higher risk (P = 0.032) of death than did the patients with CEA-positive tumors. Cox regression analysis revealed that ER was the only parameter with significant interacting effect with CEA. Multivariate, stepwise Cox regression analysis showed that CEA expression, tumor size, and nuclear grade were the only significant independent predictors of DFS, and nuclear grade and lymph node metastasis the only significant predictors of OS in the ER-positive group. The only significant independent predictor of DFS and OS in the ER-negative group was CEA. When CEA expression was stratified to ER status, patients whose tumors lacked CEA and ER had threefold higher risk of disease relapse (P = 0.002) and a 5.3-fold higher risk of death (P = 0.0001) than those with ER-positive and CEA-negative tumors. Multivariate analysis showed that the association between CEA and ER was enhanced further after compensating for other parameters with independent predictive value. CONCLUSIONS The association between CEA and ER was the most important independent predictor of a subgroup of patients (CEA-negative, ER-positive) with the most favorable prognosis. The results imply that the association of several tumor markers may provide tumor profiles with superior predictive value than a single parameter.