One centennial discussion in linguistics concerns whether languages, or linguistic systems, are, essentially, homogeneous or rather show "structured heterogeneity." In this contribution, the question is addressed whether and how sociolinguistically defined systems (or 'varieties') are to be distinguished in a heterogeneous linguistic landscape: to what extent can structure be found in the myriads of language variants heard in everyday language use? We first elaborate on the theoretical importance of this 'variety question' by relating it to current approaches from, among others, generative linguistics (competing grammars), sociolinguistics (style-shifting, polylanguaging), and cognitive linguistics (prototype theory). Possible criteria for defining and detecting varieties are introduced, which are subsequently tested empirically, using a self-compiled corpus of spoken Dutch in West Flanders (Belgium). This empirical study demonstrates that the speech repertoire of the studied West Flemish speakers consists of four varieties, viz. a fairly stable dialect variety, a more or less virtual standard Dutch variety, and two intermediate varieties, which we will label 'cleaned-up dialect' and 'substandard.' On the methodological level, this case-study underscores the importance of speech corpora comprising both inter- and intra-speaker variation on the one hand, and the merits of triangulating qualitative and quantitative approaches on the other.