Abstract: A novel registration method of preoperative CT images and intra-operative X-ray images is proposed for minimally invasive image-guided spine surgery.Firstly,an Iterative Closest Point algorithm is applied for coarse registration.Secondly,gradient feature is extracted from image.Based on the principle of projective transformation,the fine registration is performed by finding the maximum CT gradient projections.The correct rate of registration was of 92％ and78％ respectively on simulated data and clinical phantom research.Experimental results show that proposed method is suitable for clinical application with good robustness and less degree of human intervention.