Text categorisation is a key methodological aspect of research into discourse and translation patterns. It relies on the conceptualisation of the object of study and has significant implications for corpus compilation and analysis. In the case of legal translation research, it raises long-debated questions on the legal nature of text types and the boundaries between them. After a brief review of recurrent issues and models of legal text classification, focus is placed on three representative international settings of institutional legal translation (the European Union, the United Nations and the World Trade Organization, including their adjudicative bodies) in order to illustrate the challenges of defining the contours and internal structure of a broad area of translation practice. A cyclical multidimensional approach based on the legal contextualisation of institutional functions is applied to the mapping of three parallel corpora, one per organisation. The resulting subdivisions form an overarching matrix composed of four primary categories and several subcategories of interrelated genres. This categorisation, of unprecedented comparative coverage in institutional translation, serves to highlight the implications of more expansive or restrictive approaches to surveying legal translation.