In this paper we explain how merging of ontologies is captured by the pushout construction from category theory, and argue that this is a very natural approach to the problem. We study this independent of a specific choice of ontology representation language, and thus provide a sort of blueprint for the development of algorithms applicable in practice. For this purpose, we view category theory as a universal ‘meta specification language” that enables us to specify properties of ontological relationships and constructions in a way that does not depend on any particular implementation. This can be achieved since the basic objects of study in category theory are the relationships between multiple ontological specifications, not the internal structure of a single knowledge representation. Categorical pushouts are already considered in some approaches to ontology research (Jannink et al. 1998; Schorlemmer, Potter, and Robertson 2002; Goguen 2005; Kent 2005) and we do not claim our treatment to be entirely original. Still we have the impression that the potential of category theoretic approaches is by far not exhausted in todays ontology research. For our particular case the treatment will focus on the ontology merging, for which we will give both intuitive explanations and precise definitions. This reflects our belief that, at the current stage of research, it is not desirable to fade out the mathematical details of the categorical approach completely, since the interfaces to current techniques in ontology research are not yet available to their full extent. We will also keep this treatment rather general, not narrowing the discussion to specific formalisms — this added generality is one of the strengths of category theory. A long version of this paper with a tutorial character is available from the first author’s homepage.