Bat algorithm is a novel swarm intelligent version by simulating the bat seeking behavior. However, due to the poor exploitation, the performance is not well. Therefore, a new variant named bat algorithm with Gaussian walk is proposed in which the original uniform sample manner is replaced by one Gaussian sample. In this paper, we want to further investigate the performance with Levy sampling manner. To test its performance, three other variants of bat algorithms are employed to compare, simulation results show our modification is superior to other algorithms.