A widely debated issue in philosophy of logic concerns the possibility of an inferentialist account of classical logic. Many proposals to show that classical logic satisfies the requirements of inferentialist semantics (such as harmony) demand to modify the ordinary natural deduction rules. In this paper, we try to explain why the ordinary natural deduction rules for classical logic are not harmonious and therefore not directly justifiable within an inferentialist framework. We show however that an indirect justification of classical logic, passing through negative translation, can be acceptable from an inferentialist point of view.