We have been developed automated preoperative planning system for total hip arthroplasty. One problem in preprocessing of input data is to reconstruct segmented bone models, as most part of the process depend on manual operation and take amount of time. In this paper, we describe the evaluation semi-automated segmentation method of the femur from 3D CT images using multi-level statistical shape model in this system. We evaluated this method by measuring motion of hip joint and comparing manual segmentation. We applied this method to 5 cases, and confirmed max error is 3 degree. This result showed an effectiveness of propose method.