In this study, Artificial Neural Networks (ANN) has been used for reducing the errors of sea wave model predictions. Firstly, stand-alone PMODynamicsI model has been implemented to predict Bushehr deep-water wave characteristics. Results implies that PMODynamicsI performed better in simulating ordinary wave with height less than 1m, but it is underestimated about 75cm related to a weak  wind Global Forecasting System (GFS) forecasts during east and southeast storms. In order to increase the wave model accuracy, a MLP ANN system consists of three layers of nodes has been defined to predict the wave model errors, which optimal selection of a number and type of input neurons among factors influence the formation of "wind waves" has helped to find the relationship between input and output in ANN to minimize model error. The combination of PMODynamicsI together with ANN technique has been improved the accuracy of the sea wave model forecast till %90 and reduced RMS error from 0.31 in stand-alone PMODynamicsI to 0.22 in combinations models. As a result of the use of combined wave and ANN systems makes accurate predictions for extreme wave about 60cm.