To devise a system for surveying the frequency of nosocomial infections in a tertiary care hospital in a developing country. Prospective selective surveillance by nurses of the charts of patients at high risk for nosocomial infections, as identified by a form completed by resident physicians. The sensitivity, specificity, and predictive value of this method of selective surveillance were compared with those for total prospective chart review by two infectious disease specialists. A university hospital in northeastern Brazil. All patients hospitalized for more than 72 hours with an identified risk factor for nosocomial infection. The ratio of nosocomial infections to 100 discharges was 13.4 and the incidence density was 11.2/1,000 patient days. The surveillance method demonstrated a sensitivity of 74% and a specificity of 99.7%. Positive predictive value was 93%, negative predictive value was 99%, and overall accuracy was 98%. This method of selective surveillance for nosocomial infections based on risk factors identified by physicians demonstrated excellent predictive value and overall accuracy and may be of use to other hospitals that lack a nursing care plan book such as the Kardex. The relative frequency of nosocomial infections significantly exceeded the rates reported from hospitals in developed countries.