Argument Ellipsis (AE) is a productive process in Hebrew, but some arguments resist ellipsis – precisely those that do not denote individuals. This constraint, a reflection of a general constraint on variables in natural language, is captured if AE sites are descended from a *pro* element that is derivationally replaced by a constituent recoverable from the antecedent. This must occur after spellout (to escape pronunciation) but prior to LF (to allow overt subextraction). The proposed analysis integrates novel data as well as recent findings from studies of AE in East Asian languages, and offers a new derivational path to ellipsis, which invokes neither PF-deletion nor LF-copying.