In this thesis, a set of defining properties of grammatical variables is proposed, taking particular account of the precise extent to which variants should be required to be equivalent, semantically and pragmatically. These principles are then applied in a variationist analysis of negation and interrogation in spoken French, with data from a corpus from the Somme, northern France. Computer-assisted data-handling techniques are employed, notably the Oxford Concordance Program. For the (ne) variable, a large proportion of the data is analysed in terms of preformed sequences, which strongly favour the omission of the negative particle. There is also evidence that age is the most important extra-linguistic constraint, but this is interpreted as being a case of age-grading rather than of change in progress, as has sometimes been supposed. It is suggested that the negative particle has all but disappeared from northern French vernacular styles. To check the pragmatic equivalence of variant interrogative structures, a taxonomy of communicative functions is set up, drawing from research on speech acts, conversational structure and communicative grammar. The interrogatives in the corpus are then classified in terms of this taxonomy. In Yes/No interrogatives, clitic inversion is found to be completely absent from the corpus, and the minority use of est-ce 92! is shown to be motivated by pragmatic and socio-pragmatic factors, ie it is often used when the speaker does not expect an answer. from the addressee, or to encode politeness. WH interrogatives constitute one of the most complex grammatical variables studied so far, with six variant structures occurring in the corpus, and the choice among them being constrained by a large number of linguistic, discoursal and pragmatic factors. In order to take account of the unacceptability of some structures in certain contexts, the notion of "semi-variable" tokens is proposed. This is reflected in the method of calculating each variant's relative frequencies, as these exclude those contexts where the variant would be unacceptable, or non-equivalent to the structure actually used. The productive use of clitic inversion in the corpus is seen to be minimal, and the choice of the WH-final structure (as opposed to a WH-fronted one) is shown to be motivated overwhelmingly by discoursal considerations. The female informants are found to favour the est-ce que structure (partly, again, for politeness), whereas the male speakers use rather more of a non-standard variant.