In this paper, two frequently used models of interpreter and user cooperation are described: the translation machine model and the interactive model of interpreting. The paper provides a summary of observations made on the basis of some empirical data (extracted from videotaped interpreter-mediated psychotherapeutic sessions) with regard to the adherence of participants in the communication to these models. It concludes that the translation machine model is, in essence, an ideology, but that translation machine ideology denies the interactional realities of interpreter-mediated talk, which leads to the unwarranted assumption that interpreters actually make equivalent renditions that do not need any repair strategies. The interactional approach, however, leads to the questioning of the concept of equivalence and to the use of repair strategies in the practice of interpreter-mediated talk. This ultimately leads to the mutual understanding at which this type of talk aims.