Abstract: The classic rough sets theory is based on the equivalence relation and not used to applicate incomplete information system. In order to extend the rough sets theory, an improved rough set model based on valued tolerance relation and prior probability obtained out of incomplete information system was presented. As a part of the model, an attribute reduction algorithm applicated incomplete information system was proposed. This algorithm takes the null set as the beginning, does not need core calculating, can save time and space, and simplify the solving process as wel1.Finally, the model and the attribute reduction algorithm are verified under an incomplete information system.