We prove that a social choice function is repeatedly implementable if and only if it is dynamically monotonic when the number of agents is at least three. We show how to test dynamic monotonicity by building an associated repeated game. It follows that a weaker version of Maskin monotonicity is necessary and sufficient among the social choice functions that are efficient. As an application, we show that utilitarian social choice functions, which can only be one-shot implemented with side-payments, are repeatedly implementable, as continuation payoffs can play the role of transfers. Under some additional assumptions, our results also apply when the number of agents is two.