This paper investigates the problem of running gait optimization for humanoid robot. In order to reduce energy consumption and guarantee the dynamic balance including both horizontal and vertical stability, a novel running gait generation based on opposition-based learning particle swarm optimization (PSO) is proposed which aims at high energy efficiency with better stability. In the proposed scheme of running gait generation, a population initiation policy based on domain knowledge is employed, which helps to guide searching direction guidance at the beginning. A population update mechanism based on opposition learning is proposed for speeding up the convergence and improving the diversity. Simulation results validate the proposed method.