How to credibly communicate information when there is a conflict of interest is a question of practical importance in many domains. Although social science scholarship suggests costly signaling can improve credibility in theory, there is limited empirical evidence on how people actually behave toward costly signals in practice. In particular, how credibility perception evolves as a function of signal costs remains unknown. I conduct experiments to test whether – and precisely how – costly signals change credibility in the two canonical mechanisms of costly signaling: sinking costs and tying hands. The results reveal that the credibility function for sunk-cost signaling is a logistic curve. This behavioral pattern is remarkably stable even when the experiment is replicated in China. In contrast, the credibility function for tied-hands signaling appears to be weakly logarithmic in that signals start to gain credibility in the low-cost region. These results uncover for the first time the functional form of the behavioral relationship between credibility and costly signaling.