This paper argues that one should account for the endogeneity of important explanatory variables and the initial differences in technological efficiency when analyzing spatial income convergence among regions. In addition, the approach of Wooldridge (2005), who proposes a convenient solution to the initial condition problem in dynamic panels, proves to be fruitful. In a panel of 211 European regions observed from 1980 to 2005, the estimated speed of convergence is substantially higher, on average, than the legendary 2 percent found in many cross‐section studies. Moreover, it exhibits pronounced variation across regions due to factor mobility and knowledge spillovers.