Task scheduling in cloud computing is a complex problem. As it is clear, Load Balancing in clouds is a NP-Complete problem and gradient-based methods which search for an optimal solution to NP-Complete problems cannot converge to the best solution in an appropriate time. Therefore, in order to solve Load Balancing problem, evolutionary and meta-heuristic methods should be used. Thus, in this study, in order to find a solution for Load Balancing in cloud computing, Cuckoo Optimization Algorithm (COA) is used and it is compared with other methods including evolutionary and non-evolutionary algorithms. In order to prove efficiency of the method, COA is presented and simulated in Cloud-Sim simulator. Obtained results are better than results of GA and RoundRobin scheduling. Finally, it is found that the leader presented in this study gives more optimal outputs in heterogeneous (cloud) environments and user’ s request is processed in an acceptable time. Thus, agreement is achieved at service level and user’ s satisfaction is increased.