We implement a controlled, multi-site experiment to develop and test guidance principles for benefits transfers. These argue that when transferring across relatively similar sites, simple mean value transfers are to be preferred but that when sites are relatively dissimilar then value function transfers will yield lower errors. The paper also provides guidance on the appropriate specification of transferable value functions arguing that these should be developed from theoretical rather than ad-hoc statistical approaches. These principles are tested via a common format valuation study of water quality improvements across five countries. While this provides an idealised tested, results support the above principles and suggest directions for future transfer studies.