This study aimed to estimate and evaluate the under-recording of AIDS cases that evolved to death. From 1991 to 1995, the Mortality Information System recorded 9,213 adult deaths due to AIDS, related by name, date of birth, and date of death to 15,505 AIDS recorded by the Disease Data Registration System from 1982 to 1996. This procedure showed that 51.9% of AIDS deaths recorded in the period were not recorded as AIDS cases as they should have been. Univariate and multivariate analysis showed that women had a greater chance of under-recording than men (OR = 1.27). A lower probability of proper recording was observed in individuals with less schooling, and illiterates had a two-fold chance of not being properly recorded, as compared to individuals with college degrees. Deaths that occurred in health facilities classified as private units had a greater probability of not being recorded than those occurring elsewhere (mainly those classified as public reference units) (OR = 2.58). Deaths occurring in the city of Rio de Janeiro had a greater probability of under-recording than those in other cities (OR = 2.20).