Abstract: In the literature, innovation is the outcome of spatial interactions between the innovative structure of a district and its knowledge structure. Using data mining and spatial exploratory statistics, we find evidence of concentration of innovations in regions where one finds the necessary knowledge to develop the innovating activity. Then, we estimate a model with spatial interaction, thereby taking into account the technological spillovers. The core of the model is an innovation function, where the level of innovation depends upon the local level of research, of the level of research in neighboring areas, and of the local industrial specialization. Our results show that there is a polarized spatial structure for innovations as well as for research activities. Moreover, we find that the emergence of patents is locally influenced by the global research activity, as well as by the specific structure of local research activities. In addition, we have taken into account that proximity increases the chances for the diffusions of ideas, even though the space concentration of French patents does not seem to be seriously affected by research spillovers.