peer reviewed ; Deflection has been rampaging in the history of Dutch, but in this long-term process, the genitive proves to be remarkably resilient. One of the tenacious relics is the partitive genitive of the type ‘iets moois’ (‘something beautiful’). The construction has drawn a lot of scholarly attention (Schultink 1962: 79-80; Kester 1996: 199-224; Broekhuis Strang 1996; Hoeksema 1998; Booij 2010: 223-228; Broekhuis 2013: 420-426), but these accounts are mostly based on limited amounts of data, often gathered through introspection, yielding an oversimplifying description of the ins and outs of the construction. In our talk, we argue that the realisation of the genitival -s morpheme is subject to variation, and is influenced by both morphosyntactic factors and ‘lectal’ (i.e. regiolectal, register …) factors. Taking a usage-based perspective, we argue that the variation cannot be circumvented in providing an adequate description (see also Bybee 2010; Geeraerts Kristiansen, forthc.). Our study is the first to bring to bear substantial empirical data to the theoretical discussion on the construction at issue. We looked at corpus data (ConDiv, see Grondelaers et al. 2000), and used mixed-effects logistic regression to assess the multifactorial nature of the realisation of the partitive genitive. Our results show that the following variables have an impact on the realisation of the -s: the type of adjective, the regional variety, the register, the type of quantifier used, and the frequency of the phrase. We also looked at interaction effects and at the effect of the random factor in our model. Probing deeper into the relevant variables, we come up with statistical evidence for what can be called ‘constructional contamination’, in the form of (i) analogical influence of neighbouring constructions and (ii) influence of the regiolectal provenance of the lexical types involved. What we mean by (i) is that historically unrelated constructions that bear a superficial resemblance to the construction at issue (in our ...