Due to the exposed nature of wireless links, the communication of wireless networks is vulnerable to jammers. And because the jammer models are usually unknown to communication users, particularly in military confrontation applications, how to ensure maintain communication under different jamming is an active research topic. In this paper, we take the anti-jamming task of cognitive radio as a Markov decision process and propose an anti-jamming method based on Q-learning. The method aim to learn an efficient policy for users to maximize the total channel transmission capacity in different typical jamming scenarios. The simulation results indicate that compared with the traditional anti-jamming methods, the anti-jamming method based on Q-learning can obtain better performance, and more effective against several kinds of typical jamming models.