Abstract: For the problem of dynamic channel selection in unknown distributed environment without a priori knowledge and information exchange, multi-Q learning was proposed. The dynamic channel selection problem was formulated the existence of pure strategy Nash equilibrium in graphical game was proved. At the same time, the pure strategy Nash equilibrium was proved to be global optimal solution. Simulation results show that multi-Q learning achieves high system capacity and utility of users in the graphical game are determined mainly by the degree of the node without direct relationship to the number of users.