Non-word repetition in children is a skill related to, but separable from grammatical ability. Lexical skill may bridge the gap between these two abilities. The main aim was to determine whether real-word-repetition tasks could be better as predictors of grammatical ability than non-word-repetition tasks in children with typical language. This proposal was pursued because lexical knowledge was assumed to make performance in repetition tasks more representative of other language abilities, whereas non-word-repetition tasks are heavily influenced by phonological short-term memory. In order to investigate this possibility, three repetition tasks (two real-word lists characterized by different lexical knowledge and one non-word list), were compared in three groups of three- to four-year-olds with typical language (42 children). Grammatical ability was tested through probes for third-person plural inflection and direct-object critic use. Real words were repeated more accurately than non-words and the non-words were more sensitive to Syllable length than real words. Performance on all repetition tasks was correlated with grammatical ability, but real words predicted variance in grammatical ability to a greater extent than non-words. Given the lexical information contained in real words, repetition of such words was a better predictor of grammatical ability than non-word repetition. Future research should replicate and extend these results. Tasks using real words may also have considerable clinical potential; for this reason, these tasks might also be included in studies of children with language impairment.