This paper discusses the differences between controlled medical vocabularies that are designed as external artifacts and the mental concepts that are inside users' heads and used by users for reasoning, diagnosis, and treatment. It argues that if the controlled vocabularies currently being developed for Electronic Medical Records are to be used by people at all, they should be designed with systematical considerations of the cognitive structures and processes of the users. Without such considerations, the designed vocabularies may not be usable by people, although they may or may be appropriate for machine processing.