Animals advertise aspects of individual quality in a range of situations including during agonistic contests over access to resources. The advertisement can be produced by different types of signal, which can be broadly classified into “conventional signals” and “costly signals.” In both cases, the signals are thought to be honest such that any benefits to cheating are very limited due to prohibitive costs of advertising an inaccurately high level of quality. However, some individuals may benefit by low-level cheating in systems that are otherwise honest. Here, I test this possibility for a costly signal by analyzing residuals from relationships between a key parameter of signal magnitude and two measures of fighting ability from “shell fights” in hermit crabs. Contrary to the expectations for significant use of bluffing, individuals that performed at a greater magnitude than expected for their ability were more likely to win the encounter but used fewer repetitions than those that performed below the level expected for their ability. This indicates that the level of performance is primarily driven by the cost and that the functional significance of signal residuals in this case is that they provide information on several measures of fighting ability.