Artificial neural networks are intelligent systems that have successfully been used for prediction in different medical fields. In this study, the efficiency of a neural network for predicting the survival of patients with acute pancreatitis is compared with days-of-survival obtained from patients. A three- layer back-propagation neural network was developed for this purpose. Clinical data (e.g. patient’s age, white cell count, blood sugar level, enzyme levels, etc.) were introduced as input to the network and the survival of patients was obtained as output. The weights of all layers were randomly assigned and were modified according to clinical data obtained from patient’s records. The network was trained until the error rate fell below five percent. After training, data from another set of patients (not introduced to the network before) were presented to the neural network and its output was compared to the patient’s days-of-survival. The results showed that the network significantly outperformed clinical criteria used for this purpose. Due to the importance of identifying patients with acute pancreatitis who are at a high risk of death and the inefficiency of precise predicting of clinical criteria, it could be concluded that neural networks are efficient in predicting illness severity in patients with acute pancreatitis.