In this paper, we investigate the relationship between location patterns, innovation processes and industrial clusters. In order to do this we extend a transactions costs-based classification into a knowledge-based taxonomy of clusters, along the lines suggested by a critical revision of the main assumptions underlying most of the existing literature on spatial clusters. Our arguments show that the transactions costs approach and the innovation and technological regimes framework are broadly consistent, and that real insights into the microfoundations, nature, and evolution of clusters can be provided by these classification systems.