Abstract: This paper focuses on the dissemination characteristics and the influence factors of warning in a social network. Under the guidance of complex adaptive systems and complex network theories, a dissemination model of warning by using the multi-agent software of Repast was built based on ER random network and WS small world network. A simulation was performed at different settings of social network structure and selection rules in source nodes. By analyzing the simulation results the dissemination efficiency of warning is obviously improved when the clustering coefficient is high or the source nodes are at a high level of connectivity.