In this paper we present several efficient computational procedures for defeasible reasoning while the plausible and well-defined semantics, viz. preferred models and stable models, are not given up. The proposed algorithms exploit the structural information of defeasible knowledge bases to facilitate efficient computational models.