Background: Breast cancer subtype can be classified using standard clinical markers (estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2)), supplemented with additional markers. However, automated biomarker scoring and classification schemes have not been standardized. The aim of this study was to optimize tumor classification using automated methods in order to describe subtype frequency in the African American Breast Cancer Epidemiology and Risk (AMBER) consortium. Methods: Using immunohistochemistry (IHC), we quantified the expression of ER, PR, HER2, the proliferation marker Ki67, and two basal-like biomarkers, epidermal growth factor receptor (EGFR) and cytokeratin (CK)5/6, in 1381 invasive breast tumors from African American women. RNA-based (prediction analysis of microarray 50 (PAM50)) subtype, available for 574 (42%) cases, was used to optimize classification. Subtype frequency was calculated, and associations between subtype and tumor characteristics were estimated using logistic regression. Results: Relative to ER, PR and HER2 from medical records, central IHC staining and the addition of Ki67 or combined tumor grade improved accuracy for classifying PAM50-based luminal subtypes. Few triple negative cases (< 2%) lacked EGFR and CK5/6 expression, thereby providing little improvement in accuracy for identifying basal-like tumors. Relative to luminal A subtype, all other subtypes had higher combined grade and were larger, and ER-/HER2+ tumors were more often lymph node positive and late stage tumors. The frequency of basal-like tumors was 31%, exceeded only slightly by luminal A tumors (37%). Conclusions: Our findings indicate that automated IHC-based classification produces tumor subtype frequencies approximating those from PAM50-based classification and highlight high frequency of basal-like and low frequency of luminal A breast cancer in a large study of African American women.