The need of biomedical vocabularies is well known for various tasks, e.g., supporting structured data entry, decision support and electronic data exchange as well as retrieval and statistical evaluation of the data. Due to a considerable diversity of artifacts like interface terminologies, classifications and thesauri it seems to be reasonable to demand for a massive reduction and, finally, to end up with just one unique "multi-purpose world terminology". Concept-based reference terminologies like SNOMED CT might be candidates for that idea. Unfortunately, the above mentioned kinds of vocabularies cannot be replaced because of their specific purpose-dependent nature. Their mutual distinctive characteristics are outlined and compared with inherent purpose-less concept systems that are based on formal approaches like description logics. For supporting interoperability the different kinds of purpose-dependent vocabularies can and should be improved by mappings to machine-processible reference terminologies. As a side-effect, this paper may contribute to the meta-terminology in this field of medical terminology.