To assess the interobserver reproducibility for the diagnosis of medullary carcinoma of the breast (MC), 53 previously diagnosed MCs were independently assessed by six observers for growth pattern, nuclear grade (NG), inflammation, tumor margin, intraductal component, and glandular features. Tumors were reclassified as MC, atypical MC, or infiltrating ductal carcinoma according to the histopathologic criteria of Ridolfi et al. (Cancer 40:1365, 1977), Wargotz and Silverberg (Hum Pathol 19:1340, 1988), and Pedersen et al. (Br J Cancer 63:591, 1991). NG was the most reproducible parameter, and tumor margin was the least, with consensus agreement by four of six observers for 49 (92%) and 26 (49%) of cases, respectively. Utilizing the histopathologic criteria proposed by Ridolfi et al., Wargotz and Silverberg, and Pedersen et al., consensus diagnoses were achieved in 37 cases (70%), 46 cases (87%), and 51 cases (96%), respectively. A consensus diagnosis of MC in all three systems was unassociated with tumor size, axillary lymph node status or overall survival (median follow-up: 89 mo). The consensus (or better) reclassification of 44/53 (83%), 35/53 (66%), and 27/53 (51%) previously diagnosed MC as atypical MC or infiltrating ductal carcinoma by the criteria of Ridolfi et al., Wargotz and Silverberg, and Pedersen et al., respectively, suggests that MC was previously over-diagnosed. While the scheme of Pedersen et al. is the most reproducible, additional follow-up information is necessary to determine the biological significance of this classification system. To minimize these difficulties in practice, pathologists should carefully adhere to published criteria and indicate the classification system utilized.