Module-based product family design is the key support technology in design for mass customization (DFMC). In order to speed up the response to the changing customer requirements, a novel approach to predicting new configura- tional product variants is proposed based on the integration of rough set and neural network through discovering the knowl- edge for the historical configuration information. A hierarchy model of module-based product family and the corresponding formal description are presented. Additionally, the prediction framework is proposed. The methodology can reuse the discov- ered configuration rules and knowledge efficiently, as well as reduce the effort of experimental measurement to some extent. The prediction values can be regarded as the indices for the customer satisfactions. Finally, the model is verified on a newly developed refrigerator family.