The problem of endogenous stratification associated with on-site sampling has been overlooked in the Contingent Valuation Method (CVM). We find that using on-site samples of visitors overstates visitor willingness to pay (WTP) estimates relative to a household sample of visitors, and substantially overstates the unconditional population values. We provide two methods of correcting WTP of on-site samples. The uncorrected on-site sample CVM yields WTP of $132 per trip, while visitor WTP obtained from a random sample of households had a value of $66 per trip. Adaptation of choice-based sampling correction estimator to the on-site CVM data yields $73 per trip, not statistically different from the visitor value from the household survey, but significantly different from the uncorrected on-site sample value.