To explain the variation in total real hospital costs among elderly patients who died between 1984 and 1991, a cohort analytic study of the nationally representative sample of elderly subjects included in the Longitudinal Study on Aging (N = 7,527) was carried out. The cohort comprised the subset of 1,778 community-dwelling Americans who were age 70 years and older in 1984, had one or more subsequent hospital episodes, and died by 1991. Hospital charges for 1984 through 1991 were taken from the Medicare Automated Data Retrieval System. Annual hospital charges were adjusted for inflation (restated in 1984 dollars) using the hospital market basket component of the consumer price index. The natural logarithm of aggregated real charges was used in the analysis. Mean total real hospital charges were $24,956 (SD = $27,847). A standard multivariable regression model explained 9.7% of the variance in real total hospital charges. After incorporating additional measures reflecting a respondent's distribution (mean and standard deviation) of comorbidities (as measured by the number of ICD-9-CM codes [truncated at five]) during all hospitalizations in the observation window, the cause of death, and the concentration of charges in the last year of life, the explained variance increased to 29.3%.  The most important explanatory factors were the two variables controlling for the distribution of comorbidity, the variable controlling for population density, and the dichotomous variable indicating that the patient's death was related to an acute myocardial infarction. Total real hospital resources consumed by elderly decedents vary substantially. The concentration of resources consumed in the last year of a respondent's life was only marginally significant in predicting total real hospital charges over an 8-year observation window.