In a case-control study of the relationship of conjugated estrogen use to endometrial and breast cancer, we compared the availability and quality of information on risk factors from hospital charts and gynecologists' records. Of the women for whom an indication of Premarin use was recorded in either source, 19 percent would have been classified as nonusers by the hospital chart alone, a proportion that was similar for the breast (18 percent) and uterine (14 percent) cancer cases and controls (23 percent). However, for current use of Premarin, a higher proportion (28 percent) of users were identified solely through the gynecologists' records, and this proportion was even higher among controls (42 percent) than among either breast (18 percent) or uterine (15 percent) cancer cases. As a result, relative risk estimates varied according to the source of exposure date. Physicians' records also provided substantially more detail than hospital records on duration of Premarin use, especially for controls. Most demographic, medical, and reproductive variables were adequately available from the hospital charts alone. However, certain reproductive variables, such as age at first birth, presence or absence of ovaries, and age at menarche, were not adequately recorded in either source. These results suggest that gynecologists' records provide more accurate exposure data than hospital charts to determine current use of conjugated estrogens. Moreover, in the assessment of certain reproductive variables, the use of both these record-based sources may not be sufficient.