Trust is a kind of risky reliance on another person. Social scientists have offered two basic accounts of trust: predictive expectation accounts and staking (betting) accounts. Predictive expectation accounts identify trust with a judgment that performance is likely. Staking accounts identify trust with a judgment that reliance on the person's performance is worthwhile. I argue (1) that these two views of trust are different, (2) that the staking account is preferable to the predictive expectation account on grounds of intuitive adequacy and coherence with plausible explanations of action; and (3) that there are counterexamples to both accounts. I then set forward an additional necessary condition on trust (added to the staking view), according to which trust implies a moral expectation. When A trusts B to do x, A ascribes to B an obligation to do x, and holds B to this obligation. This Moral Expectation view throws new light on some of the consequences of misplaced trust. I use the example of physicians' defensive behavior/defensive medicine to illustrate this final point.