Abstract: Neural network is one of the efficient tools in the field of reservoir identification.A new method for reservoir identification is proposed which is based on the combination of the approximation principle of BP network and the technology of preprocessing.The method,first selects the proper neighbor points for every training sample according to the certain rule,then calculates the new training data with moving-weighted-average method based on the information of the selected points,and finally obtains the identifying results.This method has faster learning speed and higher prediction precision than the traditional method without preprocessing,which uses one or two hidden layers.Experimental results show that the method can rapidly,accurately identify.