Using a multi-site valuation study of groundwater quality, this paper compares the accuracy of alternative methods of transferring values from study sites to policy sites. Transferring benefit functions is more accurate than transfers of average contingent values. Relative accuracy is highly dependent on how study sites are grouped.