Based on the rough set model proposed by Pawlak, a new fuzzy C-means algorithm-rough fuzzy C-means algorithm (RFCM) is presented. The algorithm employs a new objective function which incorporates the concepts of the upper approximation and the lower approximation in rough sets, and which produces better results than Fuzzy C-mean algorithm at time complexity, clustering precision, the sensitivity to initial degree of membership matrix. The better effect can be testified by many experiments.