The study was designed to assess the differences in the quality of certification and coding practices of underlying causes of death, which the mortality statistics is based upon. The mains focus of the study was the problem of proper selection of the underlying cause of death in various diseases. In the analysis the potential impact of medical experience of the physicians and the hospitalization of patients before the death have been taken in consideration. There were 479 death certificates chosen randomly out of all certificates filled in by 240 medical doctors in 1999 in Krakow. For each death certificate the available clinical case histories for deceased persons have been collected in order that a team of medical experts could formulate their independent opinion about the underlying cause of death. From comparisons of the underlying causes of death from the death certificates with those of experts, the indices of agreement have been calculated. The best overall agreement has been found for the neoplasmatic diseases (83.1%) and the lowest for the chest diseases (30.7%). For cardiovascular diseases the overall agreement was 65.5%, however in coronary heart disease it reached 74.0% and in cerebrovascular diseases 78.7%. Generally, the better agreement has been demonstrated if the additional information about the coexistent diseases or direct causes of the death has been mentioned in the death certificates as well. The longer period of hospitalization of subjects before death was related significantly with the higher agreement indices.