We aimed to compare the predictive accuracy of consensus-derived and data-derived injury severity scores when considered alone and in combination with age and physiological status. Analyses were based on 25,111 patients. The predictive validity of each severity score was evaluated in logistic regression models predicting in-hospital mortality using measures of discrimination and calibration. Data-derived scores had consistently better predictive accuracy than consensus-derived scores in univariate models (p<0.0001) but very little difference between scores was observed in models including information on age and physiological status. Data-derived scores provide more accurate mortality prediction than consensus-derived scores when only anatomic injury severity is considered but offer little advantage if age and physiological status are taken into account.