Abstract: The objective of this paper is to use an artificial neural network(ANN) model to train the output of a third generation wave model to better forecast the significant wave heights from buoy data.After training,the agreement between the wave model's output and the buoy data generally increases,but there is still significant disagreement when the wave height is at its peak.The significant wave heights bigger than 1.5m are selected to retrain,using the same ANN model,and the resulting improvement in the forecast is obvious since the root mean square error(RMS) between the ANN output and the buoy data decrease from 0.31 m to 0.29 m.The goal of this paper is to investigate the feasibility of using an ANN to improve a wave model's numerical wave prediction so as to develop a more accurate wave forecasting system.The results show that an ANN is an useful tool for this purpose.