The revelation principle is reconsidered in the light of recent work questioning its general applicability, as well as other work on the Bayesian foundations of game theory. Im plem entation in rationalizable strategies is considered. A general ized version of the revelation principle is proposed recognizing th a t, unless agents all have dominant strategies, the outcome of any allocation mechanism depends not only upon agents' "intrinsic" types, but also upon their beliefs about other agents and their strategic behaviour. This generalization applies even if agents are "boundedly rational" in the sense of being Bayesian rational only with respect to bounded models of the game form.