Restricted by the diversity of resources and the complexity of search algorithms,current search mechanisms in peer-to-peer file sharing systems are based on file names and simple keyword matching.These mechanisms cannot recognize deeper relationships between keywords and resources;hence it cannot provide high search quality.This paper proposes a new search scheme,which is built on top of the current peer-to-peer network. It harnesses users' search behaviors and download behaviors to automatically discover the deeper relationships between keywords and resources,which is then used to improve the search quality.It has the advantages of low implementation cost,low complexity,self-evolving,and supports for semantic search.Simulations based on the Maze system show that this approach has high search hit rate and accuracy.