In this paper, we propose an approach that uses in-game reputation as a solution to the problem of cheating in massively multiplayer online games. What constitutes cheating is however quite context-specific and subjective, and there is no universal view. Thus our approach aims to adjust to the particular forms of cheating to which players object rather than deciding a priori which forms of cheating should be controlled.The main feature of our approach is an architecture and model for maintaining player-based and context-appropriate trust and reputation measures, with the integration of these into the game's ranking system. When an avatar loses reputation, our approach intervenes to reduce its ranking. It is envisaged that players will come to attach value to reputation in its own right. We also present the results of relatively large-scale simulations of various scenarios involving sequences of encounters between players, with an initial implementation of our reputation and ranking model in place, to observe the impact on cheaters (and non-cheaters).