For decades, the Random Utility Model (RUM) has been a workhorse model for such applications as valuing new products or changes in the characteristics of public goods. RUMS have sometimes been faulted along two lines. First, for including idiosyncratic errors that imply un-reasonably high values for new alternatives and unrealistic substitution patterns. Second, for involving strong restrictions on functional forms for utility. This paper shows how, starting with a RUM framework one can nonparametrically set-identify the answers to policy questions using only the Generalized Axiom of Revealed Preference (GARP). This approach recasts the RUM errors as departures from GARP, to be minimized using a minimum-distance criterion an provides another avenue for nonparametric identification of the RUM model. The paper illustrates the approach by estimating bounds on the values of ecological improvements in the Southern Appalachian Mountains, using survey data.