In order to improve the reliability of wind speed series prediction, a new associative network was constructed to predict the wind speed series with chaotic characteristics. Stored sample patterns were constructed according to the similarity measure of the volatile of the wind speed series. Utilizing the correlation information contained in the stored sample patterns, the network adopts an unsupervised learning algorithm to complete the weight training. One step or multi-step prediction of the wind speed series which have self-similarity can be completed by the associative network. Compared with the conventional forward neural network, the prediction mechanism of the associative prediction network is explicit, and the prediction result is uniqueness. The network can also give one step or multi-step prediction results simultaneously in once calculation. Simulation results show that the associative network has good prediction performance, and can be applied to predict dynamically the wind speed series.