We study the coevolution of cooperation, preferences and cooperative signals in an environment where individuals engage in a signaling-extended prisoner's dilemma. We identify a new type of evolutionary equilibrium - a transitional equilibrium - which is constituted and stabilized by the dynamic interaction of multiple Bayesian equilibria. A transitional equilibrium: (1) exists under mild conditions, and (2) can stabilize a population that is characterized by the heterogeneity of behavior, preferences, and signaling. We thereby offer an explanation for persistent regularities observed in laboratory and field data on cooperative behavior. Furthermore, this type of equilibria is least demanding with respect to differences in signaling costs between 'conditional cooperators' and 'opportunists.' Indeed, and quite surprisingly, a transitional equilibrium is consistent with 'conditional cooperators' bearing higher signaling cost in terms of fitness than 'opportunists.'