This study reports the results front an in-person comparison of contingent valuation (CV) survey response and subsequent voting behaviour on comparable propositions to increase public spending for landscape protection. A substantial proportion of CV responses were not consistent with self-reported actual voting decisions, suggesting an upward bias of stated willingness to pay. Item non-response and protest zero bids were more frequently given by those rejecting than those approving the actual proposition. Self-reported actual decisions yielded a more consistent explanatory pattern than did hypothetical choices. The official aggregate voting records show that these results cannot be explained simply by errors in self-reported votes. We conclude that referenda provide a valuable opportunity to test stated preference methods for public goods, particularly public goods with significant passive-use values. However, an important requirement is that the stated choices to be compared with the referendum decisions, unlike in previous studies, come from independent contingent valuation surveys.