Abstract: A scheme of opportunistic spectrum access based on POMDP is proposed, to avoid the interference to the primary user for the secondary user does not have the exact knowledge of which channel is occupied. The process of secondary user selecting a satisfied channel is formulated as a partial observable Markov decision process which is infinite horizon. The simulation results show that through learning from the extern world, secondary users can search for and exploit spectrum holes according to the rule of maximizing the expected rewards. The simulation results validate this model. It provides a new idea to dynamic spectrum access method.