This article shows that the empirical data that Chomsky (2008) provides in favor of feature inheritance and parallel movement are rather suspect, and that there are both empirical and conceptual reasons to reject this implementation of the idea that the formal features in C and T originate in a single head position. This does not imply, however, that the latter idea should be completely rejected. This paper argues that there are also reasons to assume that the relevant features all originate in the T-head, and that the C-position comes (or rather: may come) into existence as a result of the remerge of T as a specific instantiation of the formation of extended projections in the sense of Grimshaw (1997). We will conclude by showing that the extended projection approach is preferred to the feature inheritance approach in terms of optimization/economy given that for subject-initial sentences, CP-structures are normally harmonically bounded by TP-structures due to the fact that they invoke additional violations of *MOVE and *MERGE.