The authors compared self-reports of non-spine fractures in a cohort of elderly white women with radiologic reports and medical records. Subjects (n = 9,704) were recruited between 1986 and 1988 in Baltimore, Pittsburgh, Minneapolis, and Portland, Oregon. Eleven percent (95% confidence interval 9-13%) of self-reports of fracture were false-positive (radiographs were negative) and a total of 20% (18-23%) could not be confirmed (radiographs were negative, uncertain, or not available). Report by proxy respondent was more accurate than self-report. There were no confirmed fractures in the medical records of a random sample of 283 participants who did not report a fracture. The percent of false-positives varied by the site of the injury and was low for self-reported fractures of the shoulder or upper arm (5%; 1-13%), wrist (8%; 4-11%), and hip (11%; 5-19%), but was high for hand or finger (20%; 12-30%), rib (23%; 15-32%), and face or skull (33%; 17-54%). Having a college education was associated with increased accuracy, while a history of falls and self-reported osteoporosis were associated with decreased accuracy. The authors conclude that elderly women overreport fractures, but that self-report is relatively accurate for several important osteoporotic fractures, including those of the hip, wrist, and humerus. Self-report of "any" fracture, rib, distal extremity, and head fracture, and fractures in women with a tendency to fall or with osteoporosis should be verified by a radiologic diagnosis.