We study the impact of flood risk on house prices in France using geocoded administrative data. The risk of flooding is characterised by location in a flood zone. These flood-prone areas are defined as part of the natural flood risk prevention plan. We take into account the level of risk (low, medium, high). The administrative data provides the characteristics of the properties sold and is enhanced by information on the environment in which the transactions take place, which may have an impact on the sale price (amenities, distance to the town centre, distance to the coastline, etc.). We model the prices of these properties using hedonic models. The data cover the period from 2019 to 2023 and allow us to analyse the variation over time in the effect of flood risk. The results show that being located in a flood-prone area when the risk is high has a significant negative impact on the transaction price when the geographical area is subject to frequent flooding or has recently experienced significant flooding. The results also attest to the phenomenon of imperfect memory of flooding. We propose a theoretical framework in which the individual revises his a priori on the distribution of the flood risk on the basis of the history of associated events, and which makes it possible to take account of these imperfect memory mechanisms.