The multiuser power and channel allocation problem in cognitive radio is considered in this paper. Based on game theory, we modeled the problem into a non-cooperative game and proved that this problem is a supermodular game with the purpose to maximize the total system capacity of the network in which secondary users choose their power allocated in each channel according to their payoff function which consider both the capacity gain of themselves and the loss of the others. A distributed multiuser power and channel allocation algorithm is proposed according to our analysis. Simulation results indicate that our algorithm can achieve greater performance improvement compared with selfish power and channel allocation scheme in channel capacity.