In this paper, we present a novel distributed spectrum sharing scheme for cognitive radio which can effectively reduce the need for spectrum sensing. This is achieved by utilizing the experience of reinforcement learning. Instead of sensing all of the available spectrum arbitrarily, the scheme is designed to share the spectrum based on an optimum spectrum sharing strategy which is discovered by the agents from their interaction with the wireless communication environment. It shows that reinforcement learning enables an efficient approach of spectrum sensing. The performance of the reinforcement learning scheme is investigated and comparisons with a no learning scheme are given to illustrate the benefits of our scheme.