Memetic algorithms (MAs) which mimic culture evolution are population based heuristic searching approaches for the optimization problems. This paper presents a new memetic algorithm called shuffled particle swarm optimization (SPSO), which combines the learning strategy of particle swarm optimization (PSO) and the shuffle strategy of shuffled frog leaping algorithm (SFLA). In the proposed algorithm, the population is partitioned into several memeplexes according to the performance, and the memotypes in each memeplex evolve according to the self-learning and the learning from the best memotype of the memeplex. Furthermore, the memeplexes are shuffled and separated again to continue the evolutionary process. The combination approach contributes to the local exploration and the global exploration of SPSO. Experimental studies on the continuous parametric benchmark problems show the robustness and the global convergence property of the proposed memetic algorithm.