How to find desired resources efficiently and accurately is one fundamental challenge of any unstructured P2 P networks, which is mainly involved some difficulties in the P2 P overlay topology, data representation in peers and routing mechanism. In this paper, we address the issue of resilient routing in unstructured P2 P networks. An efficient algorithm called Query routing tree(QRT) based on maximum mutual information is proposed to improve the performance of resource searching, which has tightly associated the resource contents of peers with the logical links in P2 P network that makes the query messages forwarded more effectively in similar peers and can hit more target resources faster. Additionally, we present an optimized routing scheme with the query conditions taken into account, to obtain the optimal routing tree with the minimal information gain from the candidate tree set so as to adapt to different query types more flexibly. The simulation results show the proposed QRT can reduce the search cost more effectively and maintain higher targets hit rate than existing typical algorithms such as Flooding, k-RW and APS. Finally, our optimized scheme is also proved to conduct high searching performance with nicer self-adaptability and scalability in unstructured P2 P networks. How to find recommended resources efficiently and accurately is one fundamental challenge of any unstructured P2 P networks, which is mainly involved some difficulties in the P2 P overlay topology, data representation in peers and routing mechanism. In this paper, we address the issue of resilient routing in unstructured P2 P networks. An efficient algorithm called Query routing tree (QRT) based on maximum mutual information is proposed to improve the performance of resource searching, which has tightly associated the resource contents of peers with the logical links in P2 P network that makes the query messages forwarded more effectively in similar peers and can hit more target resources faster; we present an optimized routing scheme with the query conditions taken into account, to obtain the optimal routing tree with the minimal information gain from the candidate tree set so as to adapt to different query types more flexibly. The simulation results show the proposed QRT can reduce the search cost more effectively and maintain higher targets hit rate than existing typical algorithms such as Flooding, k-RW and APS. Finally, our optimized scheme is also proven to conduct high searching performance with nicer self-adaptability and scalability in unstructured P2 P networks.