This paper presents a "cluster" based search scheme in peer-to-pe er n etwork. The idea is based on the fact that data distribution in an information s ociety has structured feature. We designed an algorithm to cluster peers that ha ve similar interests. When receiving a query request,a peer will preferentially forward it to another peer which belongs to the same cluster and shares more s imilar interests. By this way search efficiency will be remarkably improved and at the same time good resilience against peer failure (the ability to withstand peer failure) is reserved.