Children learning English often omit grammatical words and morphemes, but there is still much debate over exactly why and in what contexts they do so. This study investigates the acquisition of three elements which instantiate the grammatical category of ‘inflection’ – copula be, auxiliary be and 3sg present agreement – in longitudinal transcripts from five children, whose ages range from 1;6 to 3;5 in the corpora examined. The aim is to determine whether inflection emerges as a unitary category, as predicted by some recent generative accounts, or whether it develops in a more piecemeal fashion, consistent with constructivist accounts. It is found that for each child the relative pace of development of the three morphemes studied varies significantly, suggesting that these morphemes do not depend on a unitary underlying category. Furthermore, early on, be is often used primarily with particular closed-class subjects, suggesting that forms such as he's and that's are learned as lexically specific constructions. These findings are argued to support the idea that children learn ‘inflection’ (and by hypothesis, other functional categories) not by filling in pre-specified slots in an innate structure, but by learning some specific constructions involving particular lexical items, before going on to gradually abstract more general construction types.