This paper examines the material relations of risk within a dispute about the hazards of manufacturing human insulin using gene technology, and the rôle played by the referent 'real risk' in the technical performance of risk in that dispute. It draws on recent work in science and technology studies that extends actor-network theory to examine the performance of reality in scientific practice. The multiplicity of risks in the dispute, and the links made and unmade between them, are examined. I argue that in the dispute, risks were contingently linked and separated around a referent 'real risk' that emerged within the recombinant DNA debate of the late 1970s. I contrast my account of risk with realist and relativist accounts, each of which values risk as an abstract entity. In my account, risk's value is contingent upon sets of material relations that link hazards and procedures for their minimization. Risk's realness emerges as some risks are linked and others separated, working a multiple/singular relation in an ontological politics of risk.