We study the robustness of reputation management systems against distortions in rating behavior. In a laboratory trust experiment with reputation management, we mimic a positive bias by exclusively offering the option to rate positively or to give no rating. As predicted by theoretical considerations, this bias leads to significantly less trust than a system that additionally offers a negative rating option. A system relying solely on negative ratings does not have such an adverse effect. This highlights the importance of negative ratings for the effectiveness of reputation systems.