OBJECTIVE: To classify infections according to the carrier state determined by surveillance cultures of throat and rectum, rather than by the traditional criterion of the time of onset after admission. METHODS: An observational cohort study of 3 months' duration was performed in a mixed medical---surgical intensive care unit (ICU) in a district general hospital of a subset of patients requiring mechanical ventilation for 3 days. Surveillance cultures from throat and rectum were obtained on admission to the ICU and then twice weekly to distinguish carriage of potentially pathogenic microorganisms (PPM) brought in by the patient from microorganisms acquired during the ICU stay. RESULTS: Out of the total population of 104 patients, 21 patients were enrolled over 3 months. Eight patients (38%) developed 12 infections, half of which were of primary endogenous pathogenesis and caused by Haemophilus influenzae, Candida albicans and Pseudomonas aeruginosa carried by the patients on admission. The remaining six were of secondary endogenous pathogenesis and caused by Acinetobacter baumannii and Pseudomonas aeruginosa acquired in the unit. CONCLUSIONS: Traditional classifications of hospital infection are challenged. If the traditional 48-h cut-off point was used, then 9 of 12 cases (75%) of infection would have been classified as nosocomial, whereas using the method based on the carrier state, 50% of all infections were caused by microorganisms carried by the patient on admission to the ICU. Moreover, we believe that the distinction between primary endogenous, secondary endogenous and exogenous is valid because these three types of infection each require different control methods.