I consider variation within French and its status in speakers’ mental grammars. I start with Massot's (2008) claim that, within relevant grammatical units, speakers in contemporary metropolitan France do not combine socio-stylistically marked L and H features, and his explanation of this in terms of diglossia (Ferguson, 1959), that is, the idea that speakers possess two (in this case massively overlapping but not identical) ‘French’ grammars which co-exist in their minds: one (français démotique, FD: acquired early, well, and in a naturalistic environment) comprises one set of grammatical features which generate unmarked forms and the marked L forms; the other (français classique tardif, FCT: learnt later, often unreliably, in a more formal context and under the influence of literacy) comprises a (partially) different set of grammatical features which generate the same unmarked forms as well as the marked H forms. Speakers switch between FD and FCT but do not use them both simultaneously, at least not within the context of an individual clause. While Massot's claim is controversial (see Coveney, 2011), I provisionally accept that it is correct, and move on to consider his explanation. I review instances of variation for which I suggest Massot's model needs to be revised in order to account for the phenomenon of surface forms which can be generated by both putative grammars, and which are therefore superficially part of the overlap, but which have a different linguistic status in each and underlyingly are not therefore part of any overlap. I then reconsider Massot's two-grammar hypothesis, raising issues surrounding the extent of the overlap between them, the nature of the differences between them, and their respective statuses in the minds of speakers. I suggest that in view of their massive overlap, their non-random differences, and their contrasting cognitive statuses, it does not make sense to view both FD and FCT as autonomous grammars. Rather, I suggest that only FD is an autonomous grammar. Since the differences between FD and FCT are instantiations of naturally occurring developments usually conceptualised in terms of cyclic grammaticalisation and renewal (the L features of FD are innovations with respect to the H features of FCT), I suggest that FCT should be seen as a dependent grammatical ‘bolt on’ which encodes its conservatism in an abstract and economical way.