We examined pathology records to determine the accuracy of reports of breast cancer among 125 first-degree relatives by 414 breast cancer cases and 429 population-based controls. Cases who reported only one affected relative tended to overreport the occurrence of breast cancer to a slightly larger extent (4%) than controls (3%). One-third of reports by cases with at least two affected relatives were erroneous, compared with none in the control group. Both cases and controls committed absolute errors of approximately 2 years when reporting the age at diagnosis of relatives. These results indicate that some recall bias is introduced when self-reports of breast cancer occurrence in first-degree relatives are used in case-control studies.