A growing body of developmental evidence suggests that the cognitive abilities that enable the expression and comprehension of communicative intentions – so-called pragmatic abilities – which underlie language use and understanding, develop early. However, a puzzling feature of pragmatic development is children's difficulties with non-literal uses of language (e.g., ‘I love you so much I could eat you up!’). I outline a research program that aims to provide input to a novel theoretical account of pragmatic development that resolves this developmental puzzle. Rather than investigating different types of non-literal language use in isolation, we should adopt a global perspective on children's pragmatic difficulties. On the basis of experimental evidence from children's comprehension of metonymy and irony, I hypothesise that children's growing sensitivity to sense conventions impedes their pragmatic reasoning with non-literal uses during the preschool years. According to this hypothesis, children's ‘literalism’ does not result from poor pragmatic abilities, but arises because attending to conventions – and to sense conventions in particular - serves an important function at a particular stage of language and social learning. The aim is to open a new direction for empirical research into the development of non-literal uses of language, and pragmatic development more generally.