The relationship between diabetes mellitus and cancer risk was investigated using data from an integrated series of case-control studies conducted in Northern Italy between 1983 and 1992. Cases were 9,991 patients with incident, histologically confirmed neoplasms below age 75, including 181 cancers of the oral cavity and pharynx, 316 of the oesophagus, 723 of the stomach, 828 of the colon, 498 of the rectum, 320 of the liver, 58 of the gall bladder, 362 of the pancreas, 242 of the larynx, 3,415 of the breast, 726 of the endometrium, 971 of the ovary, 125 of the prostate, 431 of the bladder, 187 of the kidney, 208 of the thyroid, 80 Hodgkin's lymphomas, 200 non-Hodgkin's lymphomas and 120 multiple myelomas. Controls were 7,834 subjects in hospital for acute, non-neoplastic, non-metabolic, non-hormone-related disorders. A history of diabetes was reported by 5.1% of male and 5.4% of female controls. Significantly elevated relative risks (RRs) among subjects with diabetes were observed for cancers of the liver [RR = 2.8, 95% confidence interval (CI) 2.0-3.9], pancreas (RR = 2.1, 95% CI 1.5-2.9) and endometrium (RR 3.4, 95% CI 2.7-4.3). After allowance for obesity and education as well as age and sex, the RRs were 3.0 for liver, 2.3 for pancreas, and 2.8 for endometrium. Diabetic subjects had no elevated risk for any of the other cancer sites considered. For liver and endometrial cancer the RRs remained elevated up to 10 years after diagnosis of diabetes (RR 2.6 and 2.0 respectively), while the RR for pancreatic cancer declined from 3.2 in the first 5 years after diagnosis of diabetes to 2.3 from 5 to 9 years and to 1.3 (95% CI 0.7-2.3) 10 or more years since diagnosis. This suggests that the relationship between diabetes mellitus and liver and endometrial cancer is probably real, while that with pancreatic cancer is compatible with diabetes being an early symptom of the disease, or at least of preneoplastic lesions.