This paper contributes to the literature of flood risk capitalization and farmland valuation by taking the advantage of boundary discontinuities around flood zone to control for unobserved heterogeneities in estimating flood risk premium using hedonic regression. With a model controlling for building structures attached to farmland, distance to city center, flood exposure, farmland acreage, and municipality related time-invariant fixed effects, a spatial difference-in-differences framework based on boundary discontinuities is proposed. The proposed approach finds qualitatively different results than a conventional hedonic price model with municipality fixed effects. On average, in the study region (Lancaster County of Pennsylvania), the conventional hedonic price model suggests that there is a 6.66% (or $868.76/acre, in 2015 USD) value reduction due to potential flood risk in the 100-year (1% annual chance of flooding) FEMA (Federal Emergency Management Agency) flood zone. The proposed spatial difference-in-differences model, however, finds that the impact on farmland values due to the potential flood risk is insignificant with the same data. This suggests that estimating flood risk premium using a conventional hedonic price model directly may suffer from serious biases due to unobserved spatial heterogeneities. The proposed spatial difference-in-differences approach has an advantage of controlling for unobserved heterogeneities that are continuously distributed across the boundary of floodplain.