This paper tests a series of prominent hypotheses regarding the determinants of per-capita income using a novel spatial econometric approach to control for spillovers among neighboring countries and for spatially correlated omitted variables. We use simultaneous equations to identify alternative channels through which country characteristics might affect income, and then test the robustness of those effects. We find support for both "institutionalist" and "geographic" determinants of income. A time-varying index of institutional quality has a strong independent effect on current income, but there is also a persistent effect of geographic factors such as seasonal frost, malaria transmission, and coastal location, which influence income through their links to agricultural output, health, urbanization and trade. The data cover 95 countries across the world from 1960 through 2002, which we use to construct a pooled dataset of nine 5-year averages centered on 1960, 1965, and so on through 2000. We use both limited and full information estimators, partly based on a generalized moments (GM) estimator for spatial autoregressive coefficients, allowing for spatial error correlation, correlation across equations, and the presence of spatially lagged dependent variables.