*Abstract *Among theories of mass-count distinction, there is a debate as to whether the distinction exists at the lexico-semantic level, e.g. Link 1983, Krifka 1989, or is rather mediated through syntactic structure, e.g. Borer 2005, Acquaviva 2019. One piece of evidence in favor of the latter approach comes from classifier languages, in which all nouns seem to exhibit mass-like behavior in that they cannot combine with a numeral directly but only with the mediation of a classifier. In the present paper, we examine Tashkent Uzbek, an obligatory classifier dialect of Uzbek (Beckwith 1998, 2007). We argue that this language exhibits the mass-count distinction even in the absence of classifiers or, if they are present, in lower positions, before the classifiers are merged (at the NP/nP level). We provide a formal analysis of a range of Uzbek classifiers, showing that they are sensitive to the mass-count distinction (rather than constituting a source thereof.) Thus, we provide evidence that the mass-count distinction is lexically encoded in nouns even in classifier languages (e.g. Cheng & Sybesma 1998, Doetjes 1997, Sudo 2016).