The relationship between breast self-examination (BSE) and survival was evaluated in 2093 women with breast cancer newly diagnosed between June 1975 and February 1979. In this population self-examiners were younger, more educated, and more likely to be white, premenopausal, and married than nonexaminers. Self-examiners also tended to seek medical care more rapidly and to have earlier stages of disease at diagnosis. Five years after diagnosis, the cumulative observed survival rates from breast cancer were 76.7% among self-examiners and 60.9% among nonexaminers (P less than 0.0001). In a multivariate analysis known sociodemographic and treatment confounders accounted for 25% of the excess of breast cancer deaths among nonexaminers. Approximately half of the remaining survival differential was attributable to the more limited disease among selfexaminers. The residual association between BSE and survival may be related to uncontrolled effects of stage or other unrecognized confounders.