To the Editor. —Cancer mortality data have been used in a variety of epidemiological studies, including analyses of time trends and geographic variations, ecological correlations, casecontrol studies, and for ascertainment of outcomes in cohort studies. The validity of these studies in large part depends on the quality of information reported on the death certificates. The accuracy of death certificate reporting of cancer causes of death has been examined previously, and it has been noted that sizable numbers of reported colon cancer deaths were actually rectal in origin, and vice versa.1-3 Comparing certificate-reported causes of death with the hospital cancer diagnoses from the Second1 and Third National Cancer Surveys as reported in Percy et al2 and additional data from the Surveillance, Epidemiology, and End Results Program4 at the National Cancer Institute, we found that during the period 1947 through 1986, an increasing proportion of rectal cancer was