This paper presents a novel adaptive method to predict the next point of target on-line by means of general wavelet neural network.The scaling function and wavelet constructed by the sigmoid function are used as the active function in the network.The number of hidden units is determined by the number of wavelet decomposition and the dimension of input signal.The local connection is used in the output layer to destroy the influence of multi-dimension signal processing.It is proved that the robust property of the wavelet network is better than that of BP network in certain conditions.Simulation shows that the computational cost of this method does not increase with the complexity of the model of moving target.Moreover

the method has the higher tracking accuracy for the moving targets with variable acceleration and quick turn in a corner.