Among recent studies of pragmatic variation, two directions stand out. One locates variation in the distribution of pragmatic meanings across macro-social dimensions (gender, ethnicity, region) and seeks ways to study it (Schneider & Barron 2008). Another adopts a variationist methodology to propose pragmatic explanations for morphosyntactic variation (Cameron & Schwenter forthcoming). As their proponents admit, these are two different projects: they use different methodologies and consider different types of variation. Nevertheless, both face difficulties stemming from their ambivalent understanding of pragmatic variation as closer to either pragmatics or sociolinguistics, respectively. This paper discusses these difficulties and proposes some methodological steps that can help us move beyond them by remaining true to the pragmatic and social categories that emerge from the data itself.