The sensitivity of self-reports of cancer may differ among cultures, and not many studies have dealt with it. The authors measured the sensitivity of 615 individuals from the JPHC Study Cohort I in Japan who responded to a questionnaire in 1995 and had had a cancer registry report since 1990. Sensitivity as calculated using registered cancer as the gold standard was 0.36 for any cancer, but varied considerably by site: 0.81, 0.42, 0.41, 0.26, 0.14 and 0.08 for breast, uterus, stomach, lung, colorectum and liver, respectively. This sensitivity by site relates positively with the cancer incidence/death ratio (an index of prognosis) of each site. A false name was reported in 71% of colorectal cancer cases (namely polyp) and in 51% of stomach cases (ulcer and/or polyp). In conclusion, the sensitivity of self-reports of cancer was much lower in this cohort than in the US results except for breast cancer.