Limited evidence is available regarding the accuracy of recall for cancer screenings. We compared the self-report of stomach cancer screening with medical records in 337 residents of a rural town in northeastern Japan, who were eligible for annual screening offered by the town. Frequencies of attendance within the last 5 years were asked in a self-administered questionnaire, and determined from the records of the organization conducting the screening. When the frequencies were dichotomized as ever/never screened within 5 years, the self-report agreed substantially with medical records, but tended to overestimate the actual attendance (kappa = 0.68, sensitivity = 100%, specificity = 75.6%). Past history of gastroduodenal diseases and family history of stomach cancer were associated with the overreport. Although the population studied was highly selective (rural Japanese volunteers), review of the previous studies on mammography and Pap smear also showed that self-reports tended to be exaggerated. Self-report of cancer screening should be regarded as an overestimated indicator of its true prevalence.