Peer-to-peer (P2P) systems provide a robust, scalable and decentralized way to share and publish data. However, most existing P2P systems only provide a very rudimentary query facility; they only support equality or keyword search queries over files. We believe that future P2P applications, such as resource discovery on a grid, will require more complex query functionality. As a first step towards this goal, we propose a new distributed, fault-tolerant P2P index structure for resource discovery applications called the P-tree. P-trees efficiently evaluate range queries in addition to equality queries. We describe algorithms to maintain a P-tree under insertions and deletions of data items/peers, and evaluate its performance using both a simulation and a real distributed implementation. Our results show the efficacy of our approach.