Some animal data have raised the possibility that benzodiazepines influence the risk of selected cancers. With data collected in 1977-1991 in a US hospital-based study, the authors assessed the relation of benzodiazepine use to the risk of 11 cancers: breast (6,056 patients), large bowel (2,203), malignant melanoma (1,457), lung (1,365), endometrium (812), ovary (767), non-Hodgkin's lymphoma (382), testis (314), Hodgkin's disease (299), thyroid (111), and liver (37). Cases were compared with cancer controls (3,777 patients with other cancers) and noncancer controls (1,919 patients admitted for acute nonmalignant disorders). Relative risks were estimated for benzodiazepine use at least 4 days a week for at least 1 month, initiated at least 2 years before admission (sustained use) by multiple logistic regression with control for confounding factors. Results derived with noncancer controls were similar to those derived with cancer controls. For sustained benzodiazepine use relative to no use, relative risk estimates for all 11 cancers were compatible with 1.0 at the 0.05 level of significance. Relative risk estimates for durations of at least 5 years were also compatible with 1.0, with the exceptions of an increased estimate, of borderline statistical significance, for endometrial cancer, and a decreased estimate for ovarian cancer. Relative risk estimates both for sustained use that continued into the 2-year period before admission and for sustained use that ended up to > or = 10 years previously were compatible with 1.0, suggesting a lack of tumor promotion and no increase in the risk after a latent interval. Results were also null for diazepam, chlordiazepoxide, and other benzodiazepines considered separately. The results suggest absence of association between benzodiazepine use and the cancers considered, with the evidence stronger for the cancers with larger numbers of subjects. The similarity of results derived with cancer and noncancer controls suggests that benzodiazepines do not influence the risk of cancer as a whole.