Grouping students by "ability" is a topic of long-standing contention in English education policy, research and practice. While policy-makers have frequently advocated the practice as reflecting educational "standards", research has consistently failed to find significant benefits of "ability" grouping; and indeed has identified disadvantages for some (low-attaining) pupil groups. However, this research evidence has apparently failed to impact on practice in England. This article, contextualised by the authors' interests in education and social inequality, seeks to do two things. First, it provides a brief analysis of the existing research evidence on the impact of "ability" grouping, with particular reference to socio-economic inequality, identifying seven different explanations for the poorer progress of pupils in low sets that emerge from the literature. Second, it applies Foucaultian "analysis of discourse" to propose potential explanations for the apparent lack of traction of existing research with policy and practice, arguing that practices of "ability grouping" reflect cultural investments in discourses of "natural order" and hierarchy, with particular resonance for the discursive and political habitus of middle-class parents. The authors postulate that investing in a powerful counter-discourse of enlightenment science, illustrated via their current randomised control trial of different approaches to pupil grouping, may offer a means to challenge hegemonic discourses that underpin current classroom practice.