Abstract: Approach based on optimal cluster centers is an effective approach to radar target recognition. But its performance degrades rapidly when only a few training data are available, because badly-conditioned matrixes are generated with a few training data during the training process, and mistake appears when badly-conditioned matrixes directly participate in calculation. An improved approach based on optimal cluster centers is proposed in this paper. It solves the above problem by transforming badly-conditioned matrixes to well-conditioned matrixes, so accurate recognition results are obtained. It ensures high recognition rate when just a few training data are available. The simulation results show the efficiency of the proposed approach.