Projects are dealing with bigger stakes and facing an ever-growing complexity. In the first place, project risks have increased in number and criticality. Lists of identified project risks thus need to be broken down into more manageable clusters. Existing techniques for this are generally based on a well-known parameter such as the nature of the risk or its ownership. The limits of this approach are that project risk interactions are not properly considered. Project interdependent risks are thus often analysed and managed as if they were independent. The consequence is that there may be a lack of consideration of potential propagation through this risk network. A change may have dramatic consequences if the propagation chain is not clearly identified and/or not managed. Our objective in this paper is to propose a methodology for grouping risks so that the project risk interaction rate is maximal inside clusters and minimal outside. What we hope to achieve is a method which facilitates the coordination of complex projects which have many interrelated risks with many different risk owners. We contend that the capacity of risk owners to communicate and make coordinated decisions will be improved if they are grouped in such a way. This proposed reconfiguration of organisation is complementary to existing configurations. To do this, we first model project risk interactions through matrix representations. Then, the mathematical formulation of the problem is presented and two heuristics are introduced. A case study in the civil engineering industry (a large infrastructure public-private partnership project) is presented, which enables us to propose global recommendations, conclusions and perspectives.