The Danish National Patient Register, which includes information on all patients admitted to hospitals, has been evaluated as concerns the quality of the data included. The material examined consisted of a representative sample of 1094 patients from departments all over the country (gynaecology and obstetrics, medicine, surgery and paediatrics). Recoding of data, clinical as well as administrative, based on copies of the case records from the hospitals was carried out by two clinically working physicians (registrars). For the administrative data e.g. length of stay, satisfactory concordance was found. The validity of clinical information depended on clinical speciality and degree of diagnostic specificity. Based on the international classification the agreement on the three digit diagnostic level was better than on the five digit diagnostic level. For surgery the agreement was better than for medicine. The agreement between the diagnostic information (primary diagnosis) and the recoder in choosing primary diagnosis varied from 66-83 percent on the five digit level and between 73-89 percent on the three digit diagnostic level. If cases where the diagnosis in the registry could be regarded as an acceptable alternative were included, the agreement between the registry and recoding was 75-90%. In a subsample of the material double coding by the two coders was carried out and it was remarkable that, taken as a whole, the degree of agreement between the two coders was of the same size as between recoder and the registry. It is anticipated, however, that introduction of ICD-10 with more clear-cut rules for choice of primary diagnosis in morbidity coding will contribute to better validity and consequently improved hospital statistics.