Purpose: Neural networks (NNs) have received a great deal of attention over the last few years. They are being used for prediction and classification areas where regression models and other related statistical techniques have traditionally been used. The aim of this study was to compare the abilities of neural network (ANN) and logistic regression (LR) models to predict the risk of in- hospital mortality after coronary artery graft (CABG) surgery.Materials and Methods: A NN model was developed using the training set of 150 patients undergoing CABG surgery at Shariatie Hospital, in Tehran, in 1997. Then, it was validated in a test set of 160 patients having CABG surgery in 1998.The NN was consisted of 18 input, 4 hidden and 2 output neurons with a back-propagation algorithm (learning rate 0.12, training tolerance 0.5, sigmoid transfer function, and niaximum error 0.01).Results: At the end of the learning, the NN was able to recognize CABG, SM, HTN, and LVF variables, as important factors. The sensitivity and specificity in the training set were 100%, and in the testing set they were 99.33% and 100% respectively. In the LR model, CABG, SM, HTN variables were entered in the model. For this model, the sensitivity and specificity for the prediction of patient events (death survival) were 99% and 100% respectively.Conclusion: The prognostic accuracy of the NNs was better then LR models (ROC area 0.811 vs 0.791) and their calibration was better. The NN successfully estimated and predicted the risk of in hospital mortality rather than LR model, with regard to the selected factors.