We divide a data set on lake recreation in Wisconsin into two nonoverlapping samples, Milwaukee residents and non-Milwaukee residents. We then consider several hypothetical benefit transfers from a non-Milwaukee-based random utility model to Milwaukee residents. All transfers are for measuring water quality improvements. We consider transfers in which we assume no information on Milwaukee residents, limited (no behavioral) information, and some behavioral information. We consider simple transfers, model transfers, and updated transfers. In all cases we test the viability of the transfer by comparing it with benefits estimated from a random utility model estimated over the Milwaukee sample, which in effect is our ''true'' model. The values from the model and updated transfers typically deviated less than 10% from the true values. Confidence intervals are estimated by using the Krinsky-Robb procedure for all of the benefit measures computed from the random utility model. ¿ American Geophysical Union 1994