Self-reported cancer data from the California Teachers Study were validated by using California Cancer Registry data. The California Teachers Study cohort consists of 133,479 active and retired California teachers. In 1995-1996, data from a mailed questionnaire were linked to the California Cancer Registry data. Sensitivity and specificity of 11 types of cancer were calculated. Multivariate analyses were conducted to evaluate correlates of false-positive and false-negative reporting. Sensitivities showed great variation by cancer site. The highest sensitivities were observed for breast (96.4%) and thyroid (92.9%) cancers, whereas the lowest sensitivities were those for cervical (44.3%), endometrial (69.1%), and other skin (53.6%) cancers. The sensitivities for in situ cancers (at the time of diagnosis) were considerably lower than those for invasive cancers in about half of the cancer types surveyed. The specificities for individual cancer sites ranged from 90% to 99%; the highest were those for lung cancer, leukemia, and Hodgkin's disease (all 99.9%). The lowest specificity was for other skin cancer (90.2%). In situ stage at diagnosis and older age were significantly associated with false-positive reporting. Age and non-White race were associated with false-negative reporting. These findings suggest that the feasibility of using self-reported data without verification in epidemiologic studies of cancer varies by site.