According to the relationship between the first-order Doppler of HF radar and the wind direction over the sea surface, a new method of predicting wind direction is proposed with considering the combination of neural network and the method of the beam sampling. Using the simulated data under different input and output parameters of neural network, the wind direction is predicted when the expansion factor is odd number, the combination of the neural network and the beam sampling is utilized to predict the wind direction when the expansion factor is even number, which can eliminate the fuzziness of the wind direction. The predicted result shows a good agreement with the simulated data. From the predicted results by the neural network and the beam sampling, it is find that the error of the wind direction is about 4-6 and the mean error of the expansion factor of the wind direction is 0.26, which provides a new idea and method of predicting the wind field over the sea surface.