Does connecting a village to markets make it richer? Using Indonesia's Transmigration program, which created new villages with variation in initial connectivity, I find that otherwise similar villages that began more connected are poorer today. A one standard deviation increase in initial market access lowers GDP per capita decades later by roughly 4 percent. Higher connectivity shifts village residents out of agriculture and into local non-farm work. I argue that this reallocation lowers income because of a scale externality in agriculture. To inspect the mechanism, I use separate variation in villages' initial settlement sizes and estimate a positive agricultural scale externality. A spatial general equilibrium model of Indonesia reproduces the negative effect of market access when the scale externality is present. In counterfactuals, ignoring the externality overstates aggregate gains from road-network improvements by 18 percent and from increases in rural agricultural productivity by 64 percent. An agricultural earnings subsidy at the road program's fiscal outlay delivers roughly twice its rural gains.