In recent years, a large amount of CAD data can be created and transferred easily during the product design process, as well as for the corresponding mold design. A major challenge is how to define and calculate the similarity between free-form geometries with regard to their shape, in order to expedite the retrieval process and provide useful information to the users. Different approaches are proposed to find the level of similarity for models in B-Rep structure or polygonal mesh. Gaussian curvature values are mapped into the two-dimensional space, and a shape-based measure is obtained using statistical methods to represent the level of similarity. Then the comparison of substructures provides additional information about shape and topology. These approaches can be applied to models under various transformations, and shows potential associations among geometries belonging to a single product/mold design, or from different ones. The proposed method can also be used to develop data mining techniques for large product/mold databases.