This paper investigates a new approach for computing the inference of defeasible logic. The algorithm proposed can substantially reduced the theory size increase due to transformations while preserving the representation properties in different variants of DL. Experiments also show that our algorithm outperform traditional approach by several order of amplitudes.