To investigate the effect of breast self-examination (BSE), we compared the stages, survival, and the risk of death for 355 patients with breast cancer detected by BSE with those for 1,327 patients with breast cancer detected by chance. The early stages of the disease were found to be more common among the symptomatic breast cancer patients detected by BSE than those by chance. The 5-year overall survival rate was 94.4% for the symptomatic patients detected by BSE, and was significantly higher by 8.7% than that (85.7%) for patients detected by chance (P less than 0.001). The 10-year survival rate was 81.6% for patients detected by BSE, and 76.6% for cases detected by chance (the difference was not significant). The overall difference between the two survival curves was statistically significant by the logrank test (P less than 0.01). A multivariate analysis using the Cox proportional hazards model showed that the risk of death for patients detected by BSE was smaller by 0.570 times than that for patients detected by chance, which was statistically significant (P less than 0.05). The effect of biases inherent to BSE in the survival analysis cannot be controlled completely even after conducting multivariate analysis. These results suggest that BSE may contribute to the reduction of the risk of death through early detection of breast cancer. However, further examination should be conducted by other methods to obtain conclusive evidence.