Natural disasters, such as flood events, are expected to increase in their frequency and severity. Without additional adaptation measures, this results in higher flood risk. The information gained from flood risk models is essential in flood risk management. However, vulnerability model components in the form of depth-damage curves are often a large driver of uncertainty and vulnerability curves are rarely estimated due to a lack of empirical damage data from flood events. This study uses a unique dataset with experienced damages and the implementation of flood damage mitigation (FDM) measures on the household level, collected after the flood event in the Netherlands in 2021. Depth-damage curves that control for several hazard, exposure and vulnerability indicators are estimated and allow for an additional input in flood risk models. Previous estimates of the effectiveness of FDM measures are prone to a selection bias, as households that do, and do not, implement FDM measures systematically differ in their risk profiles. By using an Instrumental Variable (IV)-estimation, this study overcomes this selection bias and finds significant reductions in flood damage due to FDM measures.