Managing uncertainty is one of the emerging issues in Expert Systems to produce high quality decisions from even incomplete and uncertain data. The Dempster-Shafer theory offers a mathematically solid framework for fusing evidence. However, its exact computation is in exponential space complexity. In order to reduce this complexity manageable, various approximation schemes have been proposed. In this thesis, we propose a new approximation scheme. That is, we extended the considering subsets to the set of given hypotheses and their complements. And, we assigned belief to be assigned to a subset A by Dempster’s combination rule, to all the smallest supersets of A in the set of considering subsets unless A is in the hypotheses set. This scheme possesses several advantages over the other approximation schemes. The first is that our scheme yields much closer approximations than others although it requires a little additional space. The second advantage is that the concept of the belief interval is still meaningful after combining evidence because the disconfirmatory belief for a proposition is available. The performance of the proposed scheme is examined through an evaluation with a sample data.