Does the extent of cheating depend on a proper reference point? We use a real effort task that implements a two (gain versus loss frame) times two (monitored performance versus unmonitored performance) between-subjects design to examine whether cheating is reference-dependent. Our experimental findings show that self-reported performance in the unmonitored condition is significantly higher than actual performance in the monitored condition - a clear indication for cheating. However, the level of cheating is by far higher in the loss frame than in the gain frame. Furthermore, men are much more strongly affected by the framing than women.