This paper introduces neural networks technique for wind speed prediction and compares its performance with an autoregressive model. First, we studied the statistical characteristics of mean monthly and daily wind speed in Jeddah, Saudi Arabia. The autocorrelation coefficients are computed and the correlogram is found compatible with the real diurnal variation of mean wind speed. The stochastic time series analysis is found to be suitable for the description of autoregressive model that involves a time lag of one month for the mean monthly prediction and one day for the mean daily wind speed prediction. The results on a testing data indicate that the neural network approach outperforms the AR model as indicated by the prediction graph and by the root mean square errors.