We offer and test a model linking ethnic networks to global price dispersion which predicts lower price dispersion as shared ethnic populations between countries rise, effects that may reverse at higher levels as network discipline breaks down. Using Chinese, Indian and Japanese data, we find that country pairs linked by the Chinese network have significantly lower mean price dispersion. A one standard deviation increase in the size of the Chinese coethnic network lowers price dispersion by 6–33%, an effect that reverses as the network gets large. No such evidence exists for the Indian or Japanese networks.