Different approaches have been used to analyse international collaboration in science but none can fully explain its rapid growth. Using international co-authorships, we test the hypothesis that international collaboration is a self-organising network. Applying tools from network analysis, the paper shows that the growth of international co-authorships can be explained based on the organising principle of preferential attachment, although the attachment mechanism deviates from an ideal power-law. Several explanations for the deviation are explored, including that of the influence of institutional constraints on the mechanism of self-organisation.