This article reports grammatical gender attribution scores in French Williams participants (N = 28, mean chronological age = 15.1) in an experiment similar to the classic one from Karmiloff-Smith (1979) where grammatical gender was pitted against natural gender. WS participants massively opted for the masculine gender as the default one, just as MA-controls did. They differed from CA-controls, however, in that they provided fewer sex-based responses. Splitting the WS group into two subgroups did not reveal a shift to sex-based responses similar to the one found in controls. It is argued that this latter difference could plausibly be related to differences in cognitive, lexical or meta-linguistic abilities. (Contains 4 tables.)