Abstract: Blend shape animation is the method of choice for keyframe facial animation: a set of blend shapes (key facial expressions) are used to define a linear space of facial expressions. However, in order to capture a significant range of complexity of human expressions, blend shapes need to be segmented into smaller regions where key idiosyncracies of the face being animated are present. Performing this segmentation by hand requires skill and a lot of time. In this paper, we propose an automatic, physically-motivated segmentation that learns the controls and parameters directly from the set of blend shapes. We show the usefulness and efficiency of this technique for both, motion-capture animation and keyframing. We also provide a rendering algorithm to enhance the visual realism of a blend shape model.