Referendums on autonomy and secession or, in short, self-determination (SD) referendums, have been on the rise lately, particularly since the end of the Cold War. SD referendums do not always measure up to the rosy experience of the 2014 Scottish referendum. Many SD referendums cannot be considered ‘free and fair’. Worse, scholars have pointed to a number of cases where SD referendums exacerbated centre-periphery tensions and gave rise to escalatory spirals producing violence, including the oft-cited referendums held shortly before the Yugoslav civil war and East Timor’s 1999 independence referendum. Yet the existing evidence on the SD referendums-armed conflict nexus comes mainly from case study research; systematic cross-case evidence is almost completely lacking. Using large-N, statistical methods this dissertation project asks whether and under what circumstances SD referendums increase the risk of separatist armed conflict. The project makes three main contributions. At the conceptual level, the project introduces a novel typology of SD referendums that helps account for varying outcomes in terms of armed conflict. Methodologically, it takes seriously the problem of selection bias, thus developing and testing a model of the conditions under which SD referendums are held and using this model to eliminate potential confounding. Empirically, the project draws on the most complete collection of SD referendums existing and a new, encompassing data set on SD movements that allows for the capturing of essential agency dynamics at play. In combination, the project promises the most systematic treatment of the SD referendum-armed conflict nexus to date.