The objective of this study was to analyze the usefulness of the Brazilian Hospital Information System (SIH) in comparison to medical records to study factors associated with in-hospital mortality due to acute myocardial infarction (AMI). We evaluated a stratified random sample of 391 medical records (out of 1,936 hospital admissions forms) with AMI as the primary diagnosis in the city of Rio de Janeiro. Factors associated with in-hospital death were studied through logistic modeling. Models were developed directly from the SIH and from medical records. ROC curves were constructed to allow comparison of the different models. We found an AMI diagnostic confirmation = 91.7% and hospital mortality = 20.6%. The logistic model derived from medical records produced the best fit (concordance = 90.1%). Although the SIH model had a worse fit (concordance = 70.6%), the correction of keying-in and information errors using data from medical records did not significantly modify its performance. Under-recording of secondary diagnosis was high in the SIH forms and was the main limiting factor.