For the first time in the literature we estimate the contribution of spatial centrality to determine city size. We do this using archaeological data on cities of the region of Bukhara observed in the 9th CE. The unique feature of this region is that it was homogeneous in all respects (technology, amenities, climate, culture, language, religion, etc.) and has been homogeneous for the twelve centuries before the 9th CE. This homogeneity rules out confounding factors and endogeneity issues. We develop a simple general equilibrium spatial model that we estimate using the method of moments. The estimated model predicts very well the 9^th century city size thus showing that spatial centrality is the major determinant of city size. The Silk Road contributes to explaining what centrality cannot. Interestingly, the estimated on data for the same region in the 21st  century performs less well, indicating that other factors influence city size in modern economies. In  a further comparison with the 21st century, we find little evidence of the persistence of the oasis urban structure. We find instead that the centroid of the region has moved towards the economic core of the Uzbek economy, both in terms of population and location of cities. In a counterfactual exercise we use the model estimated for the 9th century to compute the counterfactual population shares of the U.S. commuting zones. As expected, the model underestimates the population share of large and central zones while overestimates the share of small and peripheral zones. This suggests that agglomeration mechanisms of modern “cities” have contributed to make large zones larger and small zones smaller. In a comparative counterfactual we estimate that centrality based on infrastructures explain about 20 percent of populations shares of U.S. commuting zones. Infrastructures also have modified the centrality with respect to ”walking paths” thereby improving ex-ante expected welfare by 8%.