Abstract: A model-retrieval approach based on shape-location codes of faces is proposed.The code for a face f is generated in two steps.First, an attributed relational graph(ARG) is constructed from a B-rep model and the ARG is transformed to a layered configuration whose center is the face f.Second, face f's shape-location code is recursively evaluated in the order of the layers, taking into account shapes of faces in each layer and their distances to the center of the configuration.Using all the face shape-location codes of a solid model, a code matrix, called shape description matrix, is constructed to represent the shape of the model.By means of shape description matrices, the representation precisions of model similarities are adjustable and model retrieval can be carried out.The experimental results show that this approach not only has high efficiency but also can retrieve models that have different levels of similarities.