Abstract. As Web services become widespread, many complex applications require service composition to cope with high scalability and heterogeneity. Centralized Web service composition approaches are not sufficient as they always limit the scalability and stability of the systems. How to efficiently compose and adapt Web services under decentralized environments has become a critical issue, and important research question in Web service composition. In this paper, a stigmergic-based approach is proposed to model dynamic interactions among Web services, and handle some issues in service composition and adaptation. In the proposed approach, Web services and resources are considered as multiple agents. Stigmergic-based self-organization among agents are adopted to evolve and adapt Web service compositions. Experimental results indicate that by using this approach, service composition can be efficiently achieved, despite dealing with incomplete information and dynamic factors in decentralized environments.