To discuss a best predictive score index in predicting death in patients with severe trauma, death warning score, and to provide a simple score for clinical use. The clinical data of 394 traumatic patients admitted to Department of Critical Care Medicine of Xi'nan Hospital of the Third Military Medical University, Daping Hospital of the Third Military Medical University, and Affiliated Hospital of Zunyi Medical College from January 2014 to December 2014 were retrospectively analyzed. The patients were divided into survival group (n = 330) and non-survival group (n = 64). The clinical data in two groups were recorded as following: gender, age; respiratory rate, heart rate, and systolic blood pressure at admission; the lowest values of serum creatinine (SCr), white blood cell count (WBC), platelet count (PLT), hematocrit (Hct), respectively, within 24 hours after admission; acute physiology and chronic health evaluation II (APACHE II) score, Glasgow coma scale (GCS) score, sequential organ failure assessment (SOFA), systemic inflammatory response syndrome (SIRS) score, injury severity score (ISS) within 24 hours of final diagnosis; the performance of emergency operation or intubation within 24 hours; incidence of sepsis, and clinical outcomes. Each observed indicator was analyzed by univariate analysis, and factors leading to death were further analyzed by logistic regression. Risk factors of severe trauma patients were sorted, from which the meaningful indicators were included to calculate the warning score of death. Receiver operating characteristic curve (ROC) was plotted to evaluate the predictive value of the warning score of death in severe trauma patients. Compared with the survival group, the age in non-survival group was older (years old: 51.50 ± 18.31 vs. 45.54 ± 14.70, t = -2.456, P = 0.016); SCr was increased (µmol/L: 94.18 ± 65.51 vs. 72.42 ± 28.22, t = -2.611, P = 0.011); APACHE II score (24.30 ± 6.23 vs. 16.38 ± 6.19, t = -9.353, P < 0.001) and SOFA score were higher (7.84 ± 3.68 vs. 4.43 ± 2.75, t = -7.049, P < 0.001); and the incidence of emergency intubation [79.7% (51/64) vs. 42.7% (141/330), χ² = 29.309, P < 0.001] and sepsis was also higher [48.4% (31/64) vs. 30.3% (100/330), χ² = 18.512, P < 0.001], but PLT count (X 10⁹/L: 112.75 ± 59.85 vs. 144.12 ± 68.28, t = 3.428, P = 0.001) and GCS score (6.44 ± 4.20 vs. 11.02 ± 3.93, t = 8.449, P < 0.001) were significantly lower. There was no significant difference in gender, respiratory rate, heart rate, systolic blood pressure, WBC, Hct, SIRS score, ISS, or emergency operation between two groups. The indicators with statistically significant difference from the univariate analysis were further analyzed by multivariate logistic regression, and the indices included in the regression model were age ≥ 65 years [95% confidence interval (95%CI) = 0.176-1.974, P = 0.019], APACHE II score ≥ 21 (95% CI = 0.121-2.725, P = 0.032), GCS < 6 (95% CI = 0.201-3.221, P = 0.026), severe sepsis (95%CI = 0.421-2.735, P = 0.008) or septic shock (95%CI = 0.430-3.636, P = 0.013), with assigning scores of 1.0, 1.5, 1.5, 1.5, 2.0, respectively. Finally these five indicators were included into the warning score of death. It was shown by ROC curve analysis that the area under ROC curve (AUC) of warning score of death in predicting mortality in critically ill trauma patients was 0.867, which was significantly higher than that of the APACHE II score (AUC = 0.812, P = 0.022) and GCS score (AUC = 0.783, P = 0.001). When the cut-off value of warning score of death was 1.5, the sensitivity, specificity, positive predict value (+PV), negative predict value (-PV), positive likelihood ratio (+LR), negative likelihood ratio (-LR), and Youden index was 75.00%, 85.40%, 50.0%, 94.6%, 5.16, 0.29, and 0.605, respectively. Age ≥ 65 years, APACHE II score ≥ 21, GCS < 6, severe sepsis or septic shock were the risk factors of death in patients with severe trauma, and they can be considered as warning score of death in patients with severe trauma. With the score mortality can be better predicted than any other kind of score for patients with severe trauma.