In order to ensure reliability and quality, software systems must be tested. It is more difficult to test object-oriented software systems than testing procedure-oriented software systems. We have systematically proposed a new methodology, named B&W, for testing object-oriented software at the class level, which was published in ACM Transactions on Software Engineering and Methodology (Vol.7 No.3 pp.250-295). B&W methodology uses Data member Relevance Graph (DRG) to determine observational equivalence of two objects generated by executing two equivalent terms of the given class. (Each term is a syntactically legal operations sequence). If these two objects are not observational equivalent, then an error is revealed. In this paper, we refine the algorithm for constructing the DRG of the given class. This refinement enhances the effectiveness of B&W methodology greatly.