The scope of this article is to analyze the reliability of the diagnoses of hospital admissions for external causes financed by the Brazilian Unified Health System (SUS). The diagnoses for hospital admissions from the Hospital Information System (SIH-SUS) data were compared with those from the research panel created from information culled from authorizations for hospital admission (AIH) complemented with information from other sources. The reliability was analyzed in groups using crude agreement and Kappa statistics. The crude agreement was 41.1% in Londrina and 15.8% in Maringá and the highest reliability was found for groupings for self-inflicted lesions and traffic accidents in Londrina (Kappa 0.57 and 0.56) while in Maringá reliability was acceptable for self-inflicted lesions (Kappa 0.28). The low agreement with differentiated standards according to groupings of causes and according to the municipalities highlight the fact that efforts must be made to improve the quality and use of the SIH-SUS in order to understand and intervene in accidents and violence in Brazil.