Prediction of bank expenses is one of the main tools for planning in banks as well as to maintain the balance and stability of banks in general. This study aims to estimate and use a statistical model to predict the appropriate banking expenses in the period (1995- 2014) in Libya, where multiple regression method and neural networks have been applied. A
comparison between the multiple regression models and neural networks has been made using measurement standards: coefficient of determination(R2),the mean absolute percentage error (MAPE), root mean square error (RMSE), mean absolute error (MAE) and Thiel's inequality coefficient (TC). The research concluded that the best predicition model is the neural networks compared to the multiple regression model.