Relations within the Iranian branch of Indo-European have traditionally been modelled by a tree that is essentially composed of binary splits into sub-and sub-sub-branches. I will argue that this tree is an artefact of the data that happened to be available at the time, and that it is rendered outdated by new data that have come to light from contemporary and ancient languages. Moreover, the tree was methodologically problematic from the outset, both for reasons of the isoglosses on which it is based and for not taking into account distinctions such as shared innovations vs. shared archaisms and marked vs. unmarked language changes. In addition, the question arises whether the wave model or the concept of linguistic areas would be more adequate to account for the data. The discovery of a corpus of Bactrian manuscripts encourages a new approach. I argue that a sub-branch including Bactrian, Parthian and some other languages is a hypothesis that deserves to be tested; at the same time, the comparison with other Iranian languages as well as typological considerations permit to assess the role of language contact.