Many algorithms suppress skeleton associated with boundary perturbation by preventing their formation or by costly branch pruning. This work proposes a novel concept of structural and textural skeletons. The former is associated with the general shape structure and the latter with boundary perturbations. These skeletons remain disconnected to facilitate gross shape matching without the need for branch pruning. They are extracted from a multiresolution gradient vector field that is efficiently generated within a pyramidal framework. Experimental results show that these skeletons are scale and rotation invariant. They are less affected by boundary noise compared to skeletons extracted by popular iterative and non-iterative techniques.