Due to the recent improvements of 3D digitalisation, the number of 3D models available on the web is growing more and more, and there is an increasing demand for tools supporting the automatic search for 3D objects in digital archives. Existing methods for 3D shape retrieval roughly filter shape information and are mainly based on reasoning at the geometric level. In this paper, we discuss the advantages of approaching the shape matching problem through 3D graph descriptors, which describe the shape as a configuration of relevant subparts. In our approach, shapes are compared using a graph matching technique able to identify the most similar object portions. In particular, we investigate the properties of these descriptors and of our matching method in the CAD context.