A repeated game theory based distributed dynamic channel selection and power control algorithm for dynamic spectrum sharing in cognitive radio network was proposed. Without interfering with primary users, each cognitive user could select channels along with transmission power when considering its specific QoS requirements and local information. The proposed algorithm can make an efficient trade-off between system spectrum efficiency and user QoS requirement, and exhibits rapid convergence. It can meet the user-specific QoS requirement to the full extent and especially suites for heterogeneous networks.