Few studies have addressed the accuracy of self-reported cancer history, although epidemiologic studies routinely use self-reported information as the sole source of exposure or outcome data or as a criterion for exclusion from study participation. In this paper, false-negative reporting of cancer history is examined in a community-based sample by comparing interview data with tumor registry records. Subjects were participants in the 1980 New Haven Epidemiologic Catchment Area study; in 1995, cancer records (from 1935 onward) were obtained by linking the sample to the Connecticut Tumor Registry. Analyses focused on 263 individuals who had at least one tumor reported to the Connecticut Tumor Registry prior to participation in the Epidemiologic Catchment Area study. The overall rate of false-negative reporting was 39.2%. Logistic regression analysis revealed that false-negative reporting was significantly associated with non-White race, older age, increased time since cancer diagnosis, number of previous tumors, and type of cancer treatment received. In addition, false-negative reporting varied widely by cancer site, ranging from 0% for melanoma skin cancer to 83.3% for central nervous system cancers. The false-negative rate for breast cancer was 20.8%, that for colon and prostate cancers was 42.1%, and that for bladder cancer was 61.5%. Implications of these findings for prevalence estimation and future epidemiologic studies are discussed.