This paper proposes a system that practically identifies the position and orientation of hand bones from magnetic resonance (MR) volume images by registration of a bone model. Investigation of the link structure of a human hand requires the measurement of the relative movement of the bones by identifying each bone configuration among a large number of hand poses of the same subject. The proposed system aims to save total time in acquiring bone configuration with the same accuracy compared with the completely-manual segmentation. From one of the MR scanning data of different poses. a bone model is generated by manually segmenting the bone region. The initial configuration of the bone model is set interactively by an operator and the model is aligned in MR voxel data by minimizing the performance index. Experimental results show the validity oft he proposed system in terms of accuracy and required time for data-processing.