Peer-to-Peer computing has become a popular networking paradigm for file sharing, distributed computing, collaborative working, etc. The widely used unstructured Peer-to-Peer protocols mainly face two problems arecting their working efficiency: 1) inefficient flooding-based search, 2) topology mismatch between the overlay network and its underlying network. In this paper, we propose to organize nodes into a semantic overlay network called CON which is composed of special interest groups based on nodes’ contents. CON guides the content search with semantic information so that it avoids most of the flooding cost. In order to alleviate the mismatch problem, nodes in CON initialize links according to their underlay proximity. Simulation results show that our mechanism efficiently increases the query success rate and reduces the traffic cost and query latency. We also compare CON with the similar work, which illuminates that CON performs better in many aspects.