The coding of multiple sclerosis (MS) as underlying cause of death (UCD) on the death certificate provides an important epidemiologic resource for both descriptive and analytic studies. However, not all deaths among MS patients will be so coded. We investigated the effect on estimated occurrence of MS and on characteristics of MS patients when only UCD codes are used to identify cases. Of 2329 MS patients living in Los Angeles County (California) or King/Pierce Counties (Washington) in 1970, 438 had died by 1980. Only 53% of the deaths were coded to MS; 47% were attributed to other causes. Based on our comparisons, the use of only MS-coded deaths to describe decedents would: underestimate the age at MS onset; overestimate the female:male ratio; underestimate age at death; and underestimate duration of MS. Also, the percentage of MS-coded deaths decreased with lengthening duration of follow-up of these prevalent cases. The effect of using only UCD codes to report characteristics of decedents with other chronic diseases may well be similar.