Complex adaptive system (CAS) theory is used to analyze the e-commerce system and the concept of e-commerce transaction network is proposed. The e-commerce transaction network can be viewed as a complex network which is constructed by transaction behaviors between buyers and sellers.Then we give a formal definition of the e-commerce transaction network and establish a multi-agent model for e-commerce transaction network. Based on the transaction rules defined, the agent-based modeling toolkit Repast S is adopted to generate the e-commerce transaction network. The generated network is divided into three 2-mode networks that are buyer-product, seller-product and buyer-seller networks; each of them is transformed into two 1-mode networks respectively. Pajek is used to graphically show the relationships among the agents in the transaction network. The concepts of degree centrality and betweenness centrality in social network analysis are used to analyze the derived 1-mode networks, and the positions of the buyers, sellers and products in the transaction network can be determined visually. Such an approach for analyzing transaction network can be used for knowledge discovery and personalized recommendation in e-commerce.