Recent studies of lexical diversity in English-speaking children with specific language impairment (ESLI) have produced conflicting results. Differences between SLI and age-matched (AM) groups on verb types, and overall types have been reported, as have differences between SLI and language-matched (LM) groups on verb types and/or verb tokens, and noun types and noun tokens. At the same time there have been other claims that there is no difference between SLI and LM groups in terms of lexical diversity. This comparison of the lexical diversity of Cantonese-speaking children with SLI (CSLI) and their language-matched peers controlled the length of the samples and the number of tokens as the basis for comparison. There was no difference between the groups in use of verb tokens or types, but there were significant differences in noun tokens and types and 'other' open class tokens and types. Although there was no difference between the groups in the use of a specific grammatical marker (aspect markers), the way in which the CSLI children deployed these markers was severely restricted in comparison with their LM peers. A limited capacity model of language production is invoked to explain the findings.