This article analyzes the completion of death certificates related to external causes in children and adolescents residing in Recife, Pernambuco State, Brazil, from 1979 to 1995. The analysis focused on the extent to which the following variables were filled in: personal data (occupation, level of schooling, medical care, confirmation of diagnosis by post-mortem, and type of violence) and place (location and municipality of both occurrence and death). Using the chi-squared method, quantitative and/or qualitative flaws were found in most of the variables analyzed. In 1995, the "schooling" and "medical care" variables were recorded in only 5.7% and 17.9% of cases, respectively. Mismatches were observed between data on place of death and medical care, as well as between place of accident and number of accidents. The results suggest a dissociation between the objective of including the variable in the death certificate and its social function. The study provides the public sector with support for improved collecting and critical analysis of data in the mortality information system.