Behavior in trust games has been linked to general notions of trust and trustworthiness, important components of social capital. In the equilibrium of a trust game, the investor does not invest, foreseeing that the allocator would keep all of the returns. We use a human-subjects experiment to test the effects of changes to the game designed to increase cooperation and efficiency. We add a pre-play stage in which the investor receives a cheap-talk message from the allocator, observes the allocator s previous decision, or both. None of these changes alter the game s theoretical predictions. We find that allowing observation results in substantially higher cooperation and efficiency, but cheap talk has little effect.