In 1990, a total of 21 852 Japanese female residents (aged 40–59 years) from four public health centre areas completed a self-administered questionnaire, which included items about the frequency of soy consumption. Through December 1999 and 209 354 person-years of follow-up, 179 women were diagnosed with breast cancer. Cox proportional hazards regression was used to estimate the RRs and 95% CIs for breast cancer in relation to consumption of miso soup, soyfoods and estimated isoflavones. Consumption of miso soup and isoflavones, but not of soyfoods, was inversely associated with the risk of breast cancer. The associations did not change substantially after adjustment for potential confounders, including reproductive history, family history, smoking and other dietary factors. Compared with those in the lowest quartile of isoflavone intake, the adjusted RRs for breast cancer for women in the second, third, and highest quartiles were 0.76 (95% CI 0.47–1.2), 0.90 (95% CI 0.56–1.5), and 0.46 (95% CI 0.25–0.84), respectively (P_trend = 0.043). The inverse association was stronger in postmenopausal women (P_trend = 0.006).