I propose a strategy of measuring the long-run economic impact of climate change on farmland values that tackles the elusive problem of time-invariant spatially-dependent unobservables in the hedonic approach. The strategy exploits that a county’s agricultural productivity is primarily influenced by its own climate, and the fact that climate assignment appears random conditional on average county-neighborhood characteristics. Results suggest that large impacts of climate change on US agriculture seem unlikely. Findings are robust to multiple checks and cannot be attributed to measurement error. Ignoring such confounders considerably overstates long-run climate change impacts on the sector.