Multi-stage classification algorithm is proposed for the special character of ship target high resolution range profile (HRRP). Range alignment, elimination of abnormal HRRP and noncoherent average are all used to process the ship HRRP before feature extraction. Multi-stage classification algorithm is designed for different feature to contribute their predominance in different stage. Classification rules are also different to ensure the high accurate classification rate in the first 3 stages, and the BP neural networks is used to process the generalized central moments features to determine the exclusive target class in the last stage. The results of experiment with the actual measured data of 10 ships indicates that the proposed algorithm is very useful in ship target classification.