We conduct a national-scale analysis of the determinants of agricultural land values. The theoretical basis for the study is a spatial city model with stochastic returns to future land development. The empirical model of agricultural land prices is estimated with a cross-section on approximately three thousand counties in the contiguous US. The results provide evidence that option values associated with irreversible and uncertain land development are capitalized into current farmland values. For each county, we decompose the current agricultural land value into components measuring rents from agricultural production and rents from future land development.