In this work, we proposed a novel method for segmenting 3D object pointclouds into parts. The segmentation is guided based on graspabilty and is formulated based on a compositional model.