Ricardian analyses of farmland values have become a cornerstone of the literature valuing the impacts of climate change on agriculture. However, concerns about the lack of a formal econometric strategy to deal with omitted farmland characteristics have raised doubts about the identification of such impacts. This paper proposes an original method for estimating Ricardian models with plot fixed effects to control for confounding omitted variables. Specifically, we use plot-level repeat-sale data to investigate how differences in farmland prices are explained by differences in climate conditions between two sale dates in France from 1996 to 2019. We show that, in comparison to our repeat-Ricardian estimates, standard Ricardian analyses result in artificially low benefits of climate change. In particular, our repeat-Ricardian estimates indicate that hotter summers should benefit French agriculture, in complete opposition to our pooled Ricardian estimates or to the remainder of the literature. Our repeat-Ricardian results are robust to several specifications, length-definitions of climate and sub-samples. Our simulations suggest that the omitted variable bias in standard Ricardian analyses leads to an underestimation of the impacts of future climate changes of between 56% and 96%.