Abstract: Reduction of knowledge is one of the important topics in the research on rough set theory. It has been proven that computing the optimal （minimal） reduction of decision table is a NP hard problem. In the paper here, first, the significance of attributes in decision table is defined from the viewpoint of information; then, a heuristic algorithm based on mutual information for reduction of knowledge is proposed, and the complexity of this algorithm is analyzed; Finally, the experimental results show that this algorithm can find the minimal reduction for most decision tables.