In this paper, we document new evidence on the city size distribution from a lab represented by the Bukhara region from the 3rd century B.C. to the 9th. This region was homogeneous in many respects (technology, amenities, climate, culture, language, religion, etc.) yet cities had different sizes. Statistical tests show that while city locations and numbers are not distinguishable from a random draw, the population is larger in a spatially central location. The city size distribution passes the test of log normality and the rank-size relation is concave. We rationalize our findings using an estimated discrete choice location model with traveling costs and spatial centrality.