Percutaneous surgery, which accesses to inner organs via 'needle-puncture' of the skin rather than an 'open' approach, has gotten greater attention and preference in clinic. So far, using cadavers as a medium for surgeons' skill training for percutaneous release procedures remains the most common training approach, which is extremely costly in materials and untraceable in procedures. In this study we investigate a new virtual hand modeling framework for developing an ultrasound (US)-guided percutaneous release simulator. Our framework first builds a finger joint model with bones, tendons, skin surface, and joint mechanism from magnetic resonance (MR) images of a human hand. Then, we design a joint-constraint registration method to fuse MR and US hand images of the same subject. Lastly, the fused image sets are warped to match the spatial configuration of a phantom hand by a thin-plate spline (TPS) warping scheme with incremental anatomical landmarks. Experiments showed that our framework incorporating anatomical and biomechanical constraints efficiently accommodated for pose and intensity variations of an articulated structure among different modality images, thus providing a promising medium of virtual MR-US image model for practicing percutaneous release surgery.