This paper derives the fact that VP-ellipsis in Serbian is aspect-sensitive - it is allowed with certain aspectual mismatches between the antecedent and the target VP, while being restricted with others. I propose that the apparent unsystematic discrepancies in the availability of VP-ellipsis in Serbian can be accounted for under a phase-governed approach to ellipsis, whereby only phases and phasal complements can be elided, as argued in Bošković (in press (a)). However, I argue that VP-ellipsis in Serbian is even more constrained in that, beside the phasal status of the target, the phasal status of the antecedent also matters. More specifically, general parallelism requirement on ellipsis extends to the paralelism in terms of phasal status between the antecedent and the target, i.e. either both are phases or both are phasal complements.