This paper reports on an ESRC-funded project that investigated the use of corpus-based activities in a primary-school context, with children aged 8–10 years. The study explored the contributions that could be made by a corpus - comprising language written for a child audience – and a modified version of the associated software, in helping these children (all L1 English speakers) to learn about language. Activities were devised which complied with educational policies in England, so the interactions recorded often involved classification of linguistic items. The paper presents a qualitative analysis of these interactions, identifying aspects of the approach which prompted metalinguistic discourse. It suggests that, in contrast with textbooks and other reference resources, this approach may provide a flexible route into metalinguistic understanding, which maintains links with an authentic discourse.