OBJECTIVE: In 1986, data from 25,000 major trauma outcome study patients were used to relate Trauma and Injury Severity Score (TRISS) values to survival probability. The resulting norms have been widely used. Motivated by TRISS limitations, A Severity Characterization of Trauma (ASCOT) was introduced in 1990. The objective of this study was to evaluate and compare TRISS and ASCOT probability predictions using carefully collected and independently reviewed data not used in the development of those norms. DESIGN: This was a prospective data collection for consecutive admissions to four level I trauma centers participating in a major trauma outcome study. MATERIALS AND METHODS: Data from 14,296 patients admitted to the four study sites between October 1987 through 1989 were used. The indices were evaluated using measures of discrimination (disparity, sensitivity, specificity, misclassification rate, and area under the receiver-operating characteristic curve) and calibration [Hosmer-Lemeshow goodness-of-fit statistic (H-L)]. MEASUREMENTS AND MAIN RESULTS: For blunt-injured adults, ASCOT has higher sensitivity than TRISS (69.3 vs. 64.3) and meets the criterion for model calibration (H-L statistic < 15.5) needed for accurate z and W scores. The TRISS does not meet the calibration criterion (H-L = 30.7). For adults with penetrating injury, ASCOT has a substantially lower H-L value than TRISS (20.3 vs. 138.4), but neither meets the criterion. Areas under TRISS and ASCOT ROC curves are not significantly different and exceed 0.91 for blunt-injured adults and 0.95 for adults with penetrating injury. For pediatric patients, TRISS and ASCOT sensitivities (near 77%) and areas under receiver-operating characteristic curves (both exceed 0.96) are comparable, and both models satisfy the H-L criterion. CONCLUSIONS: In this age of health care decisions influenced by outcome evaluations, ASCOT's more precise description of anatomic injury and its improved calibration with actual outcomes argue for its adoption as the standard method for outcome prediction.