OBJECTIVE: The aim of this study was to demonstrate the quality of the National Mortality Information System (SIM) in a special cohort of inpatients submitted to lower limb amputation (LLA) as a result of diabetes mellitus (DM), and compare the mortality pattern by causes for the cohort and for the population base of the State of Rio de Janeiro, from 2000 to 2003. METHODS: Data were generated from the linkage of the National Hospital Admittance Authorization-SIH-SUS (2000) and SIM (2000-2003) databases. Individuals under 30 years were excluded from the study, and deaths due to violent causes were also excluded in order to analyze the mention of DM as cause of death. Analyses of causes of death were conducted both in the cohort and the population base. Quality markers of SIM in the cohort were the frequency of mention of diabetes mellitus in the death certificate and the proportion of deaths classified as non-specific causes of death, and non-specific causes of death in the population base. RESULTS: 38.0% in the special cohort (n = 977) died during the four years following LLA as a result of diabetes mellitus (DM), and 49.1% of these deaths occurred during the first year. Endocrine, nutritional metabolic (41.5%), and cardiovascular diseases (28.5%), and non-specific causes of death (8.1%) were the main underlying causes of death (UC). DM was mentioned as UC for 41.0%, as a sequential cause for 1.6%, and as contributive cause for 10.0%, among the other natural underlying causes of death besides DM on death certificates. Non-specific underlying causes of death accounted for 11.8% of death certificates in the population base. CONCLUSIONS: The high risk of death, the causes of death and the underreporting of DM on death certificates of adults with diabetes were similar to other population-based mortality studies. In the special cohort, the quality of the causes of death on certificates was considered bad. The improvement in the reliability of cause-of-death depends on the ability of physicians to register, and of healthcare facilities to codify correctly the causes of death on death certificates. The method (linkage of data and multiple causes of death) applied in this study to qualify death information was efficient and effective.