The lexical diversity of children with specific language impairment (SLI) (ages 3 years 7 months to 7 years 3 months) was compared to that of normally developing same-age peers and younger normally developing children matched according to mean length of utterance in words (MLUw). Lexical diversity was calculated from spontaneous speech samples using D, a measure that uses repeated calculations of type-token ratio (TTR) to estimate how TTR changes as the speech samples increase in size. When D computations were based on 250-word samples, developmental differences were apparent. For both children with SLI and typically developing children, older subgroups showed higher D scores than younger subgroups, and subgroups with higher MLUws showed higher D scores than subgroups with lower MLUws. Children with SU did not differ from same-age peers. At lower MLUw levels, children with SLI showed higher D scores than younger typically developing children matched for MLUw. The developmental sensitivity of D notwithstanding, comparisons using 100-utterance samples, in which the number of lexical tokens varied as a function of the children's MLUws, and comparisons between 250- and 500-word samples revealed the possible influence of sample size on this measure. However, analysis of the effect sizes using smaller and larger samples revealed that D is not affected by sample size to the degree seen for more traditional measures of lexical diversity.