This paper hypothesizes that respondents in contingent valuation surveys may form different benefit and cost levels that deviate from the levels specified by the researcher. The conceptual framework investigates potential biases based on the direction of deviations. Survey data on the restoration of wetlands in Tampa Bay show that a significant portion of the respondents deviate from the benefit and cost levels presented in the scenario. Empirical results indicate that willingness-to-pay (WTP) estimates are very sensitive to the perceived benefit and cost levels. Depending on the direction of the deviations, WTP estimates could fluctuate up to +61 percent and -82 percent, compared to the estimate from those who evaluate the scenario at the presented levels.