Twenty-six surgical breast cancer specimens were subjected to multiple estrogen receptor (ER) assays with a microsample technique. In most tumors there were large variations in ER levels in samples taken from different regions of the same tumor. After correction of individual samples for differences in carcinoma content, such variations were not abolished. Nine of the 26 breast cancers, although positive overall, were devoid of ER in some regions. For ER-positive tumors the average coefficient of variation (CV) for intratumor ER levels was 86%, with a range from 25% to 200%. This is well above the CV obtained with repeat samples of homogeneous tissue (14%). These results suggest that many ER-positive cancers may be composed of cells with a variety of ER levels. An assessment of individual intratumor ER variability may have biologic and clinical significance.