The increasingly popular immunohistochemical techniques for assay of the estrogen receptor (ER) allow localization of receptor positivity to specific cell populations. Heterogeneity of the ER in tumor cell populations may have important implications for analytic cell selection and for prognosis in ER-positive carcinomas. We studied 84 tissue blocks for level-to-level and geographic heterogeneity within level of the ER and cytokeratin by staining alternate serial sections for ER and cytokeratin. Distribution of ER and cytokeratin positivity was manually assessed. Homogenous positive staining was seen in 63 of 84 cases for ER and 71 of 84 cases for cytokeratin. Distinct geographic variability constant from level to level was seen in 7 cases for ER. In each of these cases, the cell populations stained uniformly for cytokeratin. Artifactual heterogeneity seemed to be uncommon for ER. Automated image analysis and manual ER estimation resulted in more positive cases than did the dextran-coated charcoal (DCC) technique. Interobserver correlation for the manual method seemed high, as did correlation between the manual method and image analysis. Because a majority of the immunohistochemical staining heterogeneity for ER seemed to be biologic, we believe it may mark carcinomas that are less responsive to tamoxifen and more likely to progress than would be predicted by more traditional methods of ER analysis.