The contingent valuation method for estimating willingness to pay for public goods typically adopts a single referendum question format, which is relatively statistically inefficient. As an alternative, Cooper, Hanemann, and Signorello (2002) propose the one-and-one-half bound (OOHB) format, allowing researchers to question respondents about both a lower and higher limit on project costs, thereby securing substantial gains in statistical efficiency. Using an experimental design, we find that responses to OOHB valuation questions fail crucial tests of procedural invariance. We test various competing models of observed response patterns including strategic misrepresentation of standard preferences and nonstandard models of preference formation.