This paper extends the revelation principle to environments in which the mechanism designer cannot fully commit to the outcome induced by the mechanism. We show that he may optimally use a direct mechanism under which truthful revelation is an optimal strategy for the agent. In contrast with the conventional revelation principle, however, the agent may not use this strategy with probability one. Our results provide a basic tool for studying dynamic contracting problems between a principal and a single agent. Also, we indicate that the revelation principle cannot be extended to a framework with multiple agents and limited commitment.