The article contributes to the current discourse on learning regions. It aims to identify favourable conditions for policy learning in regions. Supported by the most prominent theories on regional development and innovation, the existence of autonomous policy networks in regions is regarded as a fundamental prerequisite for generating innovations. However, previous research does not clarify how networks have to be organized and linked to the institutional framework inside and outside a region, how actors should interact in networks, and whether competitive or cooperative orientations of actors are more conducive to change. Our theoretical reasoning leads us to the conclusion that learning regions are those that manage to meet different, to a certain degree contradictory, demands regarding structures of networks and actors involved. Successful learning, the generation and implementation of new patterns of politics and new policies depends on specific structural characteristics of networks, but also on actors' cognitive dispositions, orientations, strategies and interactions. Moreover, favourable institutional settings are identified which may induce adequate network structures and support innovative policies. Finally we draw some conclusions for practical regional policy in a 'learning region'.