I suggest a unifying new approach to moral hazard. Once local incentive compatibility (L-IC) is satisfied, the problem of verifying global incentive compatibility (G-IC) is shown to be isomorphic to the well-understood problem of comparing two classes of distribution functions. The sufficient conditions for the validity of the first-order approach (FOA) provided by Rogerson and Jewitt are related to first and second order stochastic dominance, respectively. New conditions, among them one in the spirit of third order stochastic dominance, are presented. Conlon's multi-signal justifications can also be understood with this approach. New multi-signal conditions that rely on the more tractable orthant orders are provided. Even when the standard FOA is invalid, a modified FOA may be valid on the set of implementable actions. This resolves Mirrlees' famous counterexample.