Reputation systems allow to estimate the trustworthiness of entities based on their past behavior. Electronic commerce, peer-to-peer routing and collaborative environments, just to cite a few, highly benefit from using reputation systems. To guarantee an accurate estimation, reputation systems typically rely on a central authority, on the identification and authentication of all the participants, or both. In this paper, we go a step further by presenting a distributed reputation mechanism which is robust against malicious behaviors and that preserves the privacy of its clients. Guaranteed error bounds on the estimation are provided.