This paper examines the Tamazight quantifier kullu ‘all,’ which displays a Q_NP and NP_Q+clitic alternation. Two major approaches to this alternation have been proposed in other Afro-Asiatic languages. Shlonsky (1991) argues that the NP_Q-clitic pattern is derived from Q_NP by moving the NP to [Spec, QP], while Benmamoun (1999) proposes that the two patterns have basically different structures. I argue in favor of the non-uniform approach but show that both previous approaches are inadequate in dealing with kullu preceding the NP, although kullu+clitic can be explained by Benmamoun’s claim that Q+clitic is an NP adjunct. I propose that kullu is a floating quantifier attracted to an overt functional category. Given that floating quantifiers are traditionally analyzed as either NP modifiers or VP modifiers, kullu brings a new perspective to floating quantifiers and raises the issue of typological variation in quantifier floating.