In this study we investigate two reference dependence effects in a choice experiment. The first is the effect of the well-known distinction between gains and losses, the second is the effect of changing the reference value on willingness to pay (WTP) and willingness to accept (WTA). The latter has to our knowledge not been studied before. We hypothesize that there are differences between WTA and WTP, and that both value functions and their disparity are affected by the absolute value of the reference point. We test our hypotheses using a choice experiment with trade-offs between changes in flood probabilities and costs. The choice experiments elicit WTP and WTA, using two flood probability reference values, yielding four separate value functions. Our findings show that a substantial WTA–WTP disparity exists, and that this disparity increases when moving away from the reference point. Also, both WTA and WTP value functions are affected by the flood probability reference value, and the WTA–WTP disparity increases when the flood probability reference point increases. Both findings suggest that welfare effects caused by changes in public good provision depend not only on the direction of change (loss aversion), but also the reference value. Moreover, our results show that the latter effect is larger for losses than for gains. We introduce the concept of reference point updating as a possible explanation for these findings.