This paper explores testable implications of a simple model of spatial heterogeneity in earnings, a model in which this spatial heterogeneity arises both from differences in the places that people work and differences among people themselves. Using micro data from four waves of the Brazilian Census (from 1970 to 2000), we estimate a place premium of between 20% and 80% of wages after controlling for observable and unobservable differences between migrants and non migrants. We derive testable implications of three different potential causes of this relationship: causation of earnings by location, migrant selection, and externalities on non migrants. A range of evidence is incompatible with the notion that selection or externalities are principal determinants of the movement‐earnings relationship. This suggests that even within country's borders, very large productivity differences may arise and be persistent over time