The data of surveillance of nosocomial infections are used as clinical indicators of quality. Their diagnosis must therefore be accurate. One hundred medical records from surgical and intensive care patients were used to assess the accuracy of diagnosing nosocomial infections according to CDC definitions. All case histories included signs and symptoms of infection. There were 45 patients with a total of 60 nosocomial infections (18 urinary tract infections, 18 surgical site infections, 11 lower respiratory tract infections and 13 other infection types). After evaluating the medical records, 18 physicians independently decided whether or not there was a nosocomial infection and, if so, the type of underlying infection. Their results were compared to the diagnosis of two experienced physicians (gold standard). Twenty-five cases were assessed on each of four consecutive days and the results were discussed with the experienced physicians on the following morning. The discussions of the first day may thus have been influenced by the results of the following days. Overall sensitivity was 85.4% (Cl95 82.8-87.8%) and overall specificity was 97.1% (Cl95 95.8-98.0%). Among the most common types of nosocomial infections, urinary tract infections had the highest sensitivity (90.1%), followed by surgical site infections (87.3%) and lower respiratory tract infections (85.8%). The specificity for these types of infections was 98.2%, 98.9% and 99.5%, respectively. Although the method of investigation used does not fulfill the formal requirements of a validation study for diagnosing nosocomial infections, it seems useful for training in the diagnostics of nosocomial infections or for intermittent evaluation of criteria before starting surveillance activities. This method is useful for training the application of CDC definitions, particularly when the medical records are reduced to case studies that include only the basic information on determining the presence of a nosocomial infection in order to evaluate a large number of suspected cases in a short period of time.