Abstract: The bat algorithm is introduced into spectrum allocation to avoid being easily trapped into a local optimization in cognitive radio. The binary bat algorithm is modified based on the spectrum allocation model to achieve better results. Firstly, the selection strategy of the bat algorithm is changed to greedy strategy which can improve the exploitation capability at the current position. Secondly, the distribution of the best position of each bat experienced is analyzed and the statistic characteristic of the bat’s position is used to guide bats finding the optimal solution. It can enhance the speed of the convergence of bat algorithm. Finally, during the local search, the solutions are not obtained by mapping the real number to binary set but being updated directly in the discrete domain, which shortens the searching time. The simulation results show not only that the proposed algorithm is more efficient, but also the speed of convergence is faster than the conventional algorithm.