In a case-control study using several distinctive control subgroups, no increased risk of breast cancer was found in postmenopausal women who used replacement estrogens. When breast cancer patients and control patients with no breast lesions were assembled from patients who had all received mammography, the overall odds ratio was 0.5, and varied from 0.2 to 1.2 according to the patient's indication for estrogen treatment and the clinical reason for mammography. When a biopsy registry was the source of breast cancer patients and control patients with benign breast disease, the odds ratio was 0.8. In a conventional case-control comparison of breast cancer patients with hospitalized patients having other diagnoses, the odds ratio was 0.9 when specific data about exposure were used to classify patients as estrogen users, but the odds ratio was 3.3 when data about exposure were not recorded equally in the compared groups. These results show that calculations of "risk" depend on important clinical phenomena that have not previously been given suitable attention. The data also provide a possible explanation for the perplexing findings among earlier studies in which the odds ratios varied from "protective" to "causal" values. When the data from the present study are partitioned according to differences in patients' clinical characteristics and reasons for the diagnostic breast procedure, the conflicting results of earlier studies can be reconciled.