Trusting beliefs can be exploited. A trustful player who is cheated too often, should start trusting less, until her beliefs are correct. For this reason we model trust as an equilibriumphenomenon. Receivers of an offer to transact choose whether or not to cheat. Cheating entails a cost, with an idiosyncratic component and a socially determined one, decreasing with the mass of players who cheat. The model either has a unique equilibrium level of trust (the proportion of transactions not cheated on), or two — one with high and one with low trust. Differences in trust can result from different fundamentals or from different equilibria being realized. Surprisingly, under certain conditions these two alternatives are partially identifiable from an empirical point of view. Our model can be reinterpreted with the cost of cheating arising from an enforcement mechanism that punishes cheaters in a targeted way using limited resources.