Recent concepts such as ‘megaregions' and ‘polycentric urban regions' emphasize that external economies are not confined to a single urban core, but are shared among a collection of nearby and linked cities. However, empirical analyses of agglomeration and agglomeration externalities have so far neglected the multicentric spatial organization of agglomeration and the possibility of the ‘sharing’ or ‘borrowing’ of size between cities. The authors take up this empirical challenge by analyzing how different spatial structures, in particular the monocentricity–polycentricity dimension, affect the economic performance of US metropolitan areas. Ordinary least squares and two-stage least-squares models explaining labor productivity show that spatial structure matters: polycentricity is associated with higher labor productivity. This appears to justify suggestions that, compared with more monocentric metropolitan areas, agglomeration diseconomies remain relatively limited in the more polycentric metropolitan areas, whereas agglomeration externalities are to some extent shared among the cities in such an area. However, it was also found that a network of geographically proximate smaller cities cannot substitute for the urbanization externalities of a single large city.