Early detection is advocated widely as the best method to reduce the high rate of breast cancer mortality in women. The purpose of this study was to describe the detection histories of women with breast cancer and to identify factors related to the method of detection. During the period 1988-90, 3,197 women with invasive breast cancer, identified through the Wisconsin (United States) tumor registry, were interviewed. The method of cancer detection (classified as self, screening mammography, or clinical breast examination [CBE]) was analyzed using polychotomous logistic regression. Fifty-five percent (1,754/3,197) of the women found their own cancers, while 35 percent (1,122/3,197) were detected by screening mammography. Compared with self-detection, the likelihood of non-localized disease was significantly lower for tumors detected by mammography (odds ratio [OR] = 0.3, 95 percent confidence interval [CI] = 0.2-0.4) and CBE (OR = 0.6, CI = 0.4-0.7). The likelihood of cancer being detected by screening mammography increased with increasing age, education, number of prior mammograms, family history, and body mass index (weight/height2) (BMI). Women in the highest BMI quintile were 2.3 times (CI = 1.7-3.0) more likely than women in the lowest BMI quintile to have their cancers diagnosed by mammography. This association most likely results from breast tumors being more difficult to palpate in heavier women.