Information on occupation and industry obtained via an interview prior to death was compared to occupation and industry on the death certificate of 184 colon cancer patients in Utah. The data were coded blindly using a five-digit code. Overall, agreement in the five-digit codes was found for 63 per cent. The industry codes agreed for 67 per cent of the individuals, and the occupation was identical for 68 per cent. Agreement by subjective evaluation of the two data sources, disregarding the five-digit codes, was 73 per cent. There were no differences in agreement of the five-digit codes by age, sex, and county of residence. The number of years worked at the job given by interview was related to agreement. Misclassification occurred in a random manner. It is concluded that the use of death certificates to study the association of occupation and disease is most appropriate for pilot studies.