Subject description is a linguistic practice, and therefore all controlled vocabularies are built upon theories of language. However, the underlying theories are rarely discussed in greater length. The purpose of this thesis is to investigate what epistemological foundations controlled vocabularies are built upon and what consequences that has for their internal structure. The main object of interest is the category. From a rationalistic point of view categories are universal and well defined. In a pragmatic point of view categories are contextual with fluid boundaries. A pragmatic point of view has been shown by Wittgenstein and Rosch to more accurately describe how language works, therefore the standpoint of this thesis is that a controlled vocabulary that takes a pragmatic approach will be a better tool for indexing and searching. The study is done in three parts. First the theoretical literature on knowledge organisation is investigated. The findings are that there is a divide among writers who take a rationalistic and pragmatic approach. Then, guidelines and principles from IFLA, LCSH, Svenska ämnesord and MeSH are examined. In all cases the rationalist view is favored but the quirks of language are handled in different ways. In IFLA's FRSAD, the writers claim to take no theoretical stance but in reality they take a rationalistic perspective. LCSH blame the inconsistencies in the system on the many people that have worked on it over a long period of time. SAO is built with LCSH as a role model and therefore have the same problems, although they have gone towards a more pragmatic approach. MeSH, however, acknowledges that language is more complicated, and the controlled vocabulary is built on a more pragmatic foundation. Lastly, examples from the controlled vocabularies are discussed. The finding is that, in practice, controlled vocabularies work in accordance with a pragmatic perspective. This is a two years master’s thesis in Library and Information Science.