This article focuses on the controversy over whether complex numerals in English are constituents. Contra the traditional view (e.g. Hurford 1975; Greenberg 1990 [1978]), the cascading structure proposed in Ionin & Matushansky (2006, 2018) maintains that cross-linguistically a complex numeral does not form a constituent to the exclusion of the NP complement. The derivation of an additive complex numeral, e.g. twenty-two people, thus involves an underlying source form with a nominal conjunction, e.g. twenty people and two people. Based on the argumentation established in He (2015) and He et al. (2017) supporting complex numerals as constituents in Chinese and minority languages in southern China, this article first demonstrates that the non-constituency analysis is not viable for English, as the underlying forms of additive complex numerals may be ill-formed and also not semantically equivalent to the surface forms. We then offer evidence to support the constituency analysis from constituency tests and behavior of post-numeral and pre-numeral modifiers. Finally, we demonstrate that the extra mechanism of grafting, proposed by Meinunger (2015), is unnecessary for English complex numerals.