Abstract: A method based on the combination of neural networks with D-S evidence theory was proposed to recognize HRR targets. Multiple HRR images were input into Learning Vector Qualification (LVQ) neural network to estimate target type evidence, the results were fused by D-S evidence theory. Methods for feature space discretization and class evidence estimation were proposed. The origin of recognition error of neural network was analyzed. The results of emulation show that the correctness of this method is higher than that of LVQ network method obviously, the ability to counteract disturbance and noise is also raised.