This paper proposes a multi-swarm chaotic particle swarm optimization (MCPSO) to predict protein structure. MCPSO takes advantage of multiple sub-swarms with mixed chaotic search behavior to maintain the swarm diversity, which alleviates the premature convergence of the algorithm, as well as adopts adaptive chaotic search strategy guided by mean sub-swarm velocity to prompt the search efficiency of the sub-swarms. Moreover, MCPSO introduces information exchang among different sub-swarms to prompt the global convergence. The proposed algorithm was applied to some well-known benchmarks and an off-lattice protein model. The relative experimental results show that MCPSO is a robust optimization technique and is feasible for protein folding.