Because of randomness and uncertainty of the wind, it is difficult to predict wind power output accurately.In order to objectify the wind power output, we proposed a new interval prediction method. Firstly, we utilize the SOM neural network to cluster the measured data to smooth the fluctuations and reduce the noise. Secondly, we establish the classic BPNN prediction model .and record the upper and lower of each cluster.Then we regard them as several intervals to enhance the performance. Experiments are constructed on a set of measured data. The results demonstrate that our method has an excellent performance and can be reasonable applied to predict wind power output