In the prisoners' dilemma game, the only evolutionary stable strategy is defection, even though nutual cooperation yields a higher payoff. Building on a paper by Robson (1990), we introduce mutants who have the ability to send a (costly) signal, i.e., the "secret handshake," before each round of the game and to condition their actions on whether or not they observe the same signal from their opponent. A population playing the strategy "always defect" is vulnerable to secret handshake mutants who cooperate when they meet other secret handshakers and defect against tother opponents. However, these secret handshakers are in turn vulberable ot a second round of mutants who imitate the secret handshake and then defect against all opponents. But now a new group of secret handshakers with a different secret handshake can arise. Thus, play can cycle between cooperation and defection. We study the dynamics of that cycling. We show that in the limit, as the probability of mutation goes to zero, cooperation occurs on average half the time. Using simulations to study the implications of our model when the mutation probability is larger than zero, we find that it is possible for cooperation to be sustained for long periods. In general, cooperation is favored when mutual cooperation has aj large payoff advantage over mutual defection, and when the payoff advantage of unilateral defection is small. Surprisingly, however, there are cases where an increased payoff to unilateral defection actually raises the level of cooperation.