Self-reports of screening are frequently used in place of chart abstraction, particularly in outpatient settings, because they are generally less expensive and frequently provide the only information available. The authors expanded the literature on validation of self-reported mammography by including the validity of recall and by assessing covariates in a setting where women were examined more than once. In 1995, this study validated mammography use in a sample of 949 women aged 50-80 years who were members of a health maintenance organization with centralized automated records of mammographic examinations. The majority of women had had a mammogram within the previous 2 years according to self-reports and records, but self-reported rates exceeded record rates by 8.2%. Sensitivity was high (93.8%), whereas specificity was low (53.6%). The overall agreement between self-reports and records was 82.7%. The kappa value was 0.52, indicating fair agreement beyond chance. Modeling with logistic regression revealed that being a college graduate and having a first-degree relative with breast cancer were significantly associated with accurate recall. Comparison of actual time interval data revealed that disagreements consisted largely of women's underestimates of time since their last screening. These results add to knowledge about the validity of self-reported mammographic screening data in settings where women are screened more than once.