Objective: To compare the performances of the acute physiology and chronic health evaluation score (APACHE II) and the trauma-injury severity score (TRISS) on our intensive care unit (ICU) trauma patients. Setting: A six-bedded general ICU in a 500-bed community hospital. Design: Prospective, cohort study of consecutive admissions to the ICU following trauma. Subjects: One hundred and ninety severe trauma patients admitted from January 1992 to December 1993 were considered eligible. Patients in whom the hospital outcome was unknown were excluded. One hundred and sixty-two patients with known hospital outcome had complete TRISS data, while 108 of them also had complete APACHE II data. Outcome measure: Vital status at discharge from the last hospital that admitted the trauma patient. Methods: The APACHE II score and TRISS method were compared by using the Hosmer-Lemeshow goodness-of-fit statistics, receiver-operating characteristic (ROC) curve analysis and likelihood ratios. Results: The Hosmer-Lemeshow goodness-of-fit results were H=16.9, df=10, p=0.076 for TRISS, and H=18.1, df=9, p=0.033 for APACHE II. After the rearrangement from 10 to three (or two) risk classes, the χ2 goodness-of-fit were H=3.5, df=3, p=0.31 for TRISS, and H=10.7, df=2, p=0.0045 for APACHE II. The area (±SE) under the ROC curve was 0.963 (±0.019) for TRISS with 95% confidence intervals (95% CI) from 0.921 to 0.986, and for APACHE II was 0.902 (±.066) with 95% CI from 0.825 to 0.948. The difference between the two curves was non-significant (z=0.86, p=0.387), and the absolute performance gain was 12.2% with 95% CI from -10.9 to 35.3%. The positive likelihood ratio (LHR) of TRISS was 24.17 (95% CI 10.0-58.1), whereas the negative LHR was 0.14 (95% CI 0.05-0.42); for APACHE II the positive LHR was 12.7 (95% CI 3.6-44.5) and the negative LHR was 0.52 (95% CI 0.23-1.16). Conclusions: Both TRISS and APACHE II scores showed high discriminatory power, although the APACHE II calibration results were conflicting and were probably influenced by the low number of cases. Moreover, the TRISS method showed larger positive and smaller negative likelihood ratios. In conclusion, TRISS performed slightly better than APACHE II score as an audit system on our trauma population.