Abstract: Based on the stochastic particle swarm optimization algorithm that guarantees global convergence, an improved stochastic particle swarm optimization algorithm ---GAT-SPSO is proposed. During the evolution of SPSO, the best particle produced by genetic algorithm of tournament selection substitutes for the stopping particle and takes part in the evolution of next generation. Through the experiments of three multi-modal test functions, the result of simulation proves that the speed of convergence and the rate of convergence for GAT-SPSO are better than SPSO at the same dimension of search space.