The frame slotted aloha algorithm used in RFID system possesses Markov trait. A Markov chain model was built to analyze it. The mathematic relationship was obtained among the number of tags, slot number and success- identification rate. It was found out that when the tag's number exceeded the slot's number, the success-identification rate would descend rapidly. Thus an amelioration algorithm based on subset-division and dynamic matching concept was proposed. The simulation result shows that when the number of tags exceeds the one of slots, the success-identification rate of amelioration algorithm is much higher than the ones of frame slotted aloha and fixed matching method. This work provides theoretic guidance for the analysis and design of RFID system.