We present a growth model with interdependencies in the heterogeneous technological progress, physical capital and stock of knowledge that yields a growth-initial equation that can be taken to the data. We then use data on EU-NUTS2 regions and a correlated random effects specification to estimate the resulting spatial Durbin dynamic panel model with spatially weighted individual e ects. QML estimates support our model against simpler alternatives that impose a homogeneous technology and limit the sources of spatial externalities. Also, our results indicate that rich regions tend to have higher "unobserved productivity" and are likely to stay rich because of the strong time and spatial dependence of the GDP per capita. Poor regions, on the other hand, tend to enjoy "unobserved productivity" spillovers but are like to stay poor unless they increase their saving rates. Keywords: correlated random effects, Durbin model, economic growth, spatial panel data. JEL Classification: C23, O47