In this paper, the author wants to show that the science of translation has to be less prescriptive than it is; it has to observe facts, in this case real translations of phraseological units. On the other hand, the comparison not between facts of the system (langue), but of the discourse (parole) reveals a wide range of possible translations. At the same time, it demonstrates the difficulties in this special translation problem. As a result, the non-phraseological translations seem to be the greater in number, probably because of a lack of specific dictionaries, a situation which has recently begun to improve.