Choice modelling is increasingly being used to generate estimates of the value of changes in environmental quality. This is partly because of the informational efficiencies of the technique, but also because of concern about the accuracy of contingent valuation. Experimental evidence has, however, demonstrated that choice modelling tends to produce much higher valuation estimates than contingent valuation. One possible explanation for the divergence between choice modelling and contingent valuation estimates is the lack of incentive compatibility with the former technique. This potentially has several sources, including having no provision rule (eg a referendum), respondents choosing between more than two alternatives, and repeated choices. We report on the results of a series of experiments involving over 2000 subjects designed to test whether a lack of incentive compatibility is responsible for divergences in value estimates.