The increased use of damage control surgery in complex trauma patients requires accurate prognostic indicators. We compared the discriminatory capacity of commonly used trauma and intensive care unit (ICU) scores, including revised trauma score, injury severity scores, trauma score-injury severity scores, acute physiology and chronic health evaluations II, and clinical and laboratory parameters, on 83 consecutive trauma patients admitted to the ICU, undergoing damage control. Logistic regressions were built for mortality prediction within 30 days. Performances of the models were assessed in terms of discrimination and calibration. Areas under the receiver operating characteristic curve from the models were compared. Overall mortality was 38.5 per cent. A "clinical" model was constructed including ICU admission pH and hypothermia (≤ 35 C °) and the number of packed red blood cells during the first 24 hours. This model was adjusted for age and demonstrated better discrimination for mortality prediction (areas under the receiver operating characteristic curve = 0.8054) than injury severity score (P value = 0.049), abdominal trauma index (P value = 0.049), and acute physiology and chronic health evaluations II (P value = 0.001). There was no statistically significant difference in discrimination for mortality prediction between the "clinical" model and revised trauma score (P value = 0.4) and trauma score-injury severity score (P value = 0.4). We concluded that the combination of ICU admission pH and hypothermia and blood transfusions during 24 hours provided an excellent discriminatory capacity for mortality prediction in this complex patient population.