BACKGROUND: To evaluate the role of in utero exposure to metronidazole (a carcinogen in some animal models) and the risk of subsequent cancer, the authors conducted a retrospective cohort study of childhood cancer. METHODS: The cohort included 328,846 children younger than 5 years born to women enrolled in Tennessee Medicaid at any time between the last menstrual period (LMP) and the date of delivery. The cohort was identified by linking files of Tennessee Medicaid mothers ages 15-44 years and children and the children's birth and death certificates for the period January 1, 1975 through December 31, 1992. Exposure data were obtained from Medicaid pharmacy records and exposure was defined as filling a metronidazole prescription that had at least a day's supply between the 30 days prior to the LMP and the date of delivery. Study cases were cohort children diagnosed with a first primary cancer before age 5 years, identified by linking the cohort with a statewide childhood cancer database for the study period. RESULTS: Cohort members contributed 1,172,696 person-years of follow-up for analysis, with children exposed (8.1%) and not exposed (91.9%) in utero to metronidazole contributing 79,716 and 1,092,980 person-years, respectively. Of 952 children younger than 5 years in the statewide cancer database, 175 met study eligibility criteria. Of these, 42 had leukemia, 30 had central nervous system (CNS) tumors, 28 had neuroblastoma, and 75 had other cancers. Using Poisson regression modeling, children exposed to metronidazole in utero had no significant increase in adjusted relative risk (RR) for all cancers (RR: 0.81; 95% confidence interval [95% CI], 0.41-1.59), leukemia (no exposed case), CNS tumors (RR: 1.23; 95% CI, 0.29-5.21), neuroblastomas (RR: 2.60; 95% CI, 0.89-7.59), and other cancers (RR: 0.57; 95% CI, 0.18-1.82). CONCLUSIONS: The authors conclude that although there was no increase in risk for all cancers associated with in utero exposure to metronidazole, the observed increased risk for neuroblastomas, although not significant, requires further evaluation.