Clusters are supposed to enhance exchanges among firms or between firms and research institutions. This is termed “proximities effects” in this paper. There are many theories explaining proximities effects, but most of them lack a clear distinction between levels of action (individuals, social networks, firms, markets, etc.). This paper is focused on this issue, claiming that it is crucial to understand the shifts between levels of action. Embeddedness of economic activity in social networks is not viewed as a static situation, but rather as a process, with a reciprocal, decoupling. Two empirical studies on innovation in the south-west of France support this argument. One bears on the relations between academic laboratories and firms, the other on the creation of innovative companies. They show that proximity and embeddedness in local social networks are just a specific context for emergence of collaborations and access to resources in the emergence phases of new companies creation, but not necessarily a specific mode of regulation of professional or technologic relations.