The extensive literature on economic convergence has explored a wide variety of ways of measuring convergence in addition to finely tuning and improving the applicable econometric techniques. However, very few contributions analyze the relevance of the spatial level of analysis. Our hypothesis is that studying the convergence at the level of large regions (states) could conceal intraregional heterogeneity. This hypothesis is consistent with the New Economic Geography framework, which highlighted core-periphery patterns at the local level. However, this polarization mechanism may become difficult to identify with aggregated data or neoclassical dynamics operating at the same time. This paper proposes a multilevel approach to study this question. It allows the identification of possible heterogeneous local patterns of behavior within regions. It is applied to the US economy in a hierarchy of two levels: states and counties. The results show high intraclass correlation, indicating significant variance within states. An overall pattern of convergence is observed in line with previous results, although some states present internal patterns of divergence or significant changes in the rate of convergence.