Recall bias was assessed in a study of cancers reported by persons living in a community with a hazardous waste treatment facility (A) and a control community (B). The self-reported cancers were verified against medical records and pathology reports. Of the 56 cancer cases reported, 43 were in community A and 13 were in community B. The difference in incorrect reporting of neoplasms between community A and community B was 12% for neoplasms and 23% for malignancies. Before verification, there was a borderline significant association (P = 0.049) between living in community A and all self-reported cancers [odds ratio (OR) 1.88, 95% confidence interval 0.99-3.57]. The verified data showed that ORs decreased with the increasing precision of diagnosis. The effect of misclassification on the OR was an inflation by 15% for neoplasms and by 31% for malignancies. The results demonstrate the importance of verifying reported cases of disease, even a disease as well defined as cancer.