To investigate the incidence of breast cancer and prediagnostic serum levels of folate, B12, and pyridoxal 5'-phosphate (B6), we conducted a nested case-control study using resources from the Washington County (Maryland) serum bank. In 1974, 12,450 serum specimens were donated, and in 1989, 14,625 plasma specimens were donated by female residents of Washington County. One hundred ninety-five incident breast cancer cases and 195 controls were matched by age, race, menopausal status at donation, and cohort participation as well as by date of blood donation. In both cohorts and all menopausal subgroups, median B12 concentrations were lower among cases than controls. Differences reached statistical significance only among women who were postmenopausal at donation (1974 cohort, 413 versus 482 pg/ml, P = 0.03; 1989 cohort, 406 versus 452 pg/ml, P = 0.02). Among women postmenopausal at blood donation, observed associations of B12 suggested a threshold effect with increased risk of breast cancer in the lowest one-fifth compared to the higher four-fifths of the control distribution [lowest versus highest fifth: 1974 cohort, matched odds ratio = 4.00 (95% confidence interval = 1.05-15.20); 1989 cohort, matched odds ratio = 2.25 (95% confidence interval = 0.86-5.91)]. We found no evidence for an association between folate, B6, and homocysteine and breast cancer. Findings suggested a threshold effect for serum B12 with an increased risk of breast cancer among postmenopausal women in the lowest one-fifth compared to the higher four-fifths of the control distribution. These results should stimulate further investigations of potentially modifiable risk factors, such as these B-vitamins, for prevention of breast cancer.