Children with Specific Language Impairment (SLI) have language difficulties of unknown origin. Syntactic profiles are atypical, with poor performance on non-canonical structures, e.g. object relatives, suggesting a localized deficit. However, existing analyses using ANOVAs are problematic because they do not systematically address unequal variance, or fully model random effects. Consequently, a Generalised Linear Model (GLM) was used to analyze data from a Sentence Repetition (SR) task involving relative clauses. seventeen children with SLI (mean age 6;7), twenty-one Language Matched (LM) children (mean age 4;8), and seventeen Age Matched (AM) children (mean age 6;5) repeated 100 canonical and non-canonical sentences. ANOVAs found a significant Group by Canonicity interaction for the SLI versus AM contrast only. However, the GLM found no significant interaction. Consequently, arguments for a localized deficit may depend on statistical methods which are prone to exaggerate profile differences. Nonetheless, a subgroup of SLI exhibited particularly severe structural language difficulties.