Terminology variation in texts is now a well-known phenomenon, whose amount is estimated to be between 15\% to 35\%, depending on the domain, text type and kind of variants identifed. In this paper, we present different typologies of variants elaborated for application-oriented terminology enginering and synthesise them in order to derive a core of variations. We shall show that the semantic interpretation of the variation depends on the application. Even if terminology variation remains a problem in various applications, it also provides clues that could be exploited in these same applications. We illustrate what benefit they bring and conclude by highlighting future trends of variation studies.