Agents obtain more learning experiences comparing with cooperative games for one fixed agent. Two-player cooperative games are used so that each agent can learn the strategy concurrently. Reinforcement learning is used in the multi-agents based service composition method.