In this paper*, I defend the idea that the degree of cross-linguistic variability allowed by the relativized definitions of Governing Categories (henceforth GCs) for pronominals and anaphors propounded by Huang (1983), Chomsky (1986) and Aoun (1986) is not sufficient, as was already noted by Manzini & Wexler (1987) and Koster (1985, 1987). Moreover, the Basque data1 to be analyzed will also show the superiority of the latter's theory over Manzini & Wexler's. Finally, I will have to introduce slight modifications into Koster's theory, allowing for the presence of PROs to specify the maximal projections of dyadic Ns and Ps (these specifiers being submitted to the licensing condition that the Xmax which contains them be the predicate of a stative predication), so as to account for data which had already been noted in e.g. Rebuschi (1988, 1989), but had been left unexplained.