Abstract: In electric power system, economic load dispatch (ELD) is an important topic, which can not only help to build up safety and stable operation plans and prolong the service life of generating units but also can save energy and maximize the economic benefits of power enterprise. The practical ELD problem has non-smooth cost function with nonlinear constraints which make it difficult to be effectively solved. In this study, a novel global optimization algorithm, cluster evolutionary algorithm (CEA), is proposed to solve ELD problem. In CEA, a virtual cluster organization has been constructed among individuals in order to dynamically adjust the searching process of simulated evolutionary system and improve the optimization efficiency of population. In simulations, the CEA has been applied to 13 testing functions and 3 IEEE testing systems for verifying its feasibility. The experiments have shown the CEA can get high quality solutions with lesser computation cost for 13 testing functions. Compared with the other existing techniques, the proposed algorithm has shown better performance for 3 IEEE systems. Considering the quality of the solution obtained, this method seems to be a promising alternative approach for solving the ELD problem in practical power system.