A point cloud feature extraction and filtering method for position and attitude( P&A ) sensor of space non-cooperative target was presented, in order to filter the noise in raw point cloud obtained form laser P&A sensor and solve the problem that too many points taken part in the position and attitude computing wasted too much time. Then, using simulation method, the effectiveness of filtering the space rand noise and down-sample of point cloud was verified, and the robustness for target pose and Gauss measurement noise was tested. Finally, with the help of the all physical test platform for non-cooperative targets fly around, approach and capture, using the raw point cloud obtained from laser P&A sensor, the performance of the method in real position and attitude measurement was presented. The test results show that the algorithm achieves 93.1% down sampling of the original point cloud, saves 92.9% of the pose calculation time, which can effectively improve the efficiency of on-orbit data processing and the real-time performance of pose calculation.