The long-term health effects of human exposure to polybrominated biphenyls are not known. In this nested case-control study, we evaluated the association between site-specific cancer risk and serum polybrominated biphenyl levels among a Michigan cohort accidentally exposed to polybrominated biphenyls in 1973. The Michigan Department of Public Health has followed 3,899 people through 1993, among whom 195 primary cancers were identified in 187 persons. Controls were 696 randomly selected cancer-free individuals who were frequency matched to cases by sex and age (in 5-year strata). Baseline serum polybrominated biphenyl levels were measured using standard methods. We found an increasing dose-response relation for digestive system cancer risk with higher serum polybrominated biphenyl category [4-20 parts per billion (ppb), 21-50 ppb, and > 50 ppb] after adjustment for age, family cancer history, cigarette smoking, alcohol drinking, and baseline serum polychlorinated biphenyl level. Adjusted odds ratios (ORs) for each category were 8.23 [95% confidence interval (CI) = 1.27-53.3], 12.3 (95% CI = 0.80-191), and 22.9 (95% CI = 1.34-392), respectively. Univariate analysis for polybrominated biphenyl level and lymphoma risk also showed a dose-response relation, with corresponding ORs of 3.24 (95% CI = 0.24-95.9), 20.5 (95% CI = 1.51-608), and 32.6 (95% CI = 3.33-861).