We propose a partial aggregation strategy for defining the recreation sites that enter choice sets in random utility models. Under the proposal, the most popular sites and sites that will be the subject of policy analysis enter choice sets as individual sites while remaining sites are aggregated into groups of similar sites. The scheme balances the desire to include all potential substitute sites in the choice sets with practical data and modeling constraints. Unlike fully aggregate models, our analysis and empirical applications suggest that the partial aggregation approach reasonably approximates the results of a disaggregate model. The partial aggregation approach offers all of the data and computational advantages of models with aggregate sites but does not suffer from the same degree of bias as fully aggregate models. ¿ 1998 American Geophysical Union