BACKGROUND: Many investigators endeavor to predict the outcome based on admission characteristics using some established models to determine which management should be applied. However, the efficacy and applicability of the models using in the geriatric patients with severe traumatic brain injury (TBI) have not yet been evaluated. METHODS: A total of 137 geriatric severe TBI patients were enrolled in this retrospective study. Receiver operating characteristic (ROC) curves were constructed to evaluate the efficacy and usability of the International Mission for Prognosis and Analysis of Clinical Trials in TBI (IMPACT) prognostic model in evaluating the prognosis for these patients. RESULTS: The observed mortality and unfavorable outcomes at 6 months in patients with severe TBI were 54.7% and 70.8%, respectively, slightly lower than the predicted outcomes using the IMPACT model. ROC curve analysis showed areas under the curve (AUCs) for mortality of 0.76 and for unfavorable outcome of 0.80 in the Core model, of 0.76 and 0.79, respectively, in the Extended model, and of 0.73 and 0.77, respectively, in the Lab model. When expected risk of fatal outcome was >90% in any model, the true positive rate was 100%. Moreover, when the predicted risk for unfavorable outcomes was >70% in any model, the actual rate of unfavorable outcomes was >80%. CONCLUSIONS: The IMPACT prognosis calculator showed just fair discrimination when predicting the outcome of the elderly patients with severe TBI. Management decisions should be made on a case-by-case basis rather than by relying on the predicted risks identified by this model; conservative treatment might be preferable when expected risk of fatal outcome is >90%.