This article contributes to the debate over the fashionable but contested concept of ‘territorial cohesion’ in the European Union. Scholars have long recognised and traced discursive shifts in EU territorial development policies, but theoretical accounts of the drivers and parameters of such shifts are rare. This article applies the multiple streams model of agenda-setting to the territorial cohesion debate in order to explore how useful this model is in analysing and predicting the outcome of a debate. The article is structured according to the three ‘streams’ that are relevant to agenda-setting: problems, policies and politics. The analysis relies on the responses to the 2008 Green Paper on Territorial Cohesion in order to determine how politically feasible different policy solutions are. More recent developments such as the Territorial Agenda 2020 and the European Commission’s proposals for Cohesion Policy for 2014–2020 are then used to assess the predictive power of multiple streams. It is shown that the model successfully predicts the endurance of solidarity-based cohesion goals, the emergence of territorial capital as a key policy solution, and the rejection of geographical criteria for the allocation of EU Structural Funds. At the same time, the multiple streams model fails to predict the introduction of spatial planning tools into EU cohesion policy. This shows that explaining a substantial redefinition of existing policy terms requires some reference to key actors’ broader discursive strategies. The article concludes that the multiple streams model has some predictive and explanatory power; criticisms of the model as overly descriptive are exaggerated.