It is argued that the received interpretation of Frank Knight's_(1921) classic risk-uncertainty distinction-as concerning whether or not agents have subjective probabilities-constitutes a misreading of Knight. On the contrary, Knight shared the modern view that agents can be assumed always to act as if they have subjective probabilities. The authors document their contention that by uncert ainty Knight instead meant situations in which insurance markets collapse because of moral hazard or adverse selection. Knight's discussion of market failure, a lthough always informal and in placesinaccurate, was in many respects a remarka ble anticipation of the modern literature.