Purpose: This research attempted to replicate Hoffman's 2009 finding that the proportion of narrative utterances with semantic or syntactic errors (i.e., = 14% "restricted utterances") can differentiate school-age children with typical development from those with language impairment with a sensitivity of 83% and specificity of 88%. Method: Methods similar to Hoffman (2009) were used to evaluate an existing set of narratives from 16 age-matched pairs of 8- to 9-year-old children, half with known language impairment. Transcripts were segmented into T-units; a code of [RESTRICTED] was assigned to any utterance with semantic or syntactic errors. Results: A Welch's t test for independent samples revealed a statistically significant difference in the mean proportion of restricted utterances between the two groups after accommodation for an outlier with typical development. A cutoff of = 14% restricted utterances replicated Hoffman's (2009) sensitivity but not specificity. Post hoc analysis of specific error types found sensitivity and specificity rates similar to Hoffman's as well as a significant difference in means when using a proportion of sentence-internal morphosyntactic errors. Conclusion: Results support further exploration of utterance-level error coding for diagnostic purposes and future development of this approach to meet clinical assessment needs.