A new method of rough measure of the preference degree between action pairs in MCDM is presented based on the Rough Sets theory. The preference degrees of reference action pairs are quantitatively represented by the sets of intervals derived from criteria value of reference action. The comprehensive preference relations of the reference action pairs are approximated respectively by the upper and the lower sets to derive the preference information potential in reference action set in terms of decision rules with no inconsistence found in LRES method. A computational study shows that this new method achieves significant improvement over the method based on the indiscernibility relation.