Most breast carcinomas in situ (CIS) are easily categorized as ductal (DCIS) or lobular (LCIS). However, some CIS have indeterminate histologic features (CIS-IF). Prior studies have shown that E-cadherin protein expression is lost in lobular but not ductal carcinomas. Therefore, evaluation of examples of CIS-IF for E-cadherin expression by immunohistochemistry might be useful in helping to define their nature. To address this, we studied histologic features and E-cadherin expression by immunohistochemistry in 89 cases of breast CIS (28 LCIS, 33 DCIS, 28 CIS-IF). CIS-IF cases were divided into three groups based on histology: Group 1 cases had all the cytologic and architectural features typical of LCIS but showed areas of comedo-type necrosis (n = 6). Group 2 cases were CIS lesions characterized by small, uniform neoplastic cells either growing in a solid pattern with focal microacinar-like structures but with cellular dyshesion, or growing in a cohesive mosaic pattern but with occasional intracytoplasmic vacuoles (n = 17). Group 3 cases showed marked cellular pleomorphism and nuclear atypia but had the dyshesive growth pattern characteristic of LCIS (n = 5). E-cadherin staining was scored as negative, positive, or mixed (mixture of negative and positive tumor cells). All 28 cases of LCIS were E-cadherin negative, and all 33 DCIS cases were E-cadherin positive by immunohistochemistry. All cases from CIS-IF group 1 and group 3 were negative for E-cadherin, suggesting a closer kinship to LCIS than to DCIS. In contrast, CIS-IF group 2 cases were heterogeneous with respect to E-cadherin staining. Six (35.3%) cases were E-cadherin negative (more akin to LCIS), 5 (29.4%) cases were E-cadherin positive (akin to DCIS), and 6 (35.3%) cases had both E-cadherin-positive and E-cadherin-negative tumor cells, suggesting a mixed DCIS/LCIS phenotype. Our findings suggest that E-cadherin immunostaining is of value in helping to characterize breast carcinomas in situ with indeterminate features. However, validation of these observations will require clinical outcome studies.