A detailed family history was obtained from 90 ovarian cancer patients and 90 age-matched controls. Relatives who died in Alberta had their death year and cancer history verified. Year of death was accurate in 12/20 and negative history of cancer in 30/30. Cancer was reported in 229/2,544 relatives. Of these, 104 were or had been residents of Western Canada, but only 69 malignancies were confirmed through population-based cancer registries. This resulted in an error rate of 10/39 (26%) for parents, 7/31 (23%) for siblings, and 18/34 (53%) for uncles and aunts; error rate is similar for patients and controls. A "positive" history of cancer was reported by 53/90 (59%) of patients and 68/90 (76%) of the controls (chi 2 = 5.7); this difference disappears when the total number of relatives is considered: 92/1,179 (7.8%) in patients and 134/1,366 (9.8%) in controls (chi 2 = 3.3). The age of relatives should also be considered for age/sex-specific years at risk comparison. In summary, the validity of studies showing family history of cancer is questionable owing to the multiple sources of error and the lack of specificity of the methods of analysis.