Purpose: The purpose of this study was to examine (a) whether increased grammatical error rates during a standardized narrative task are a more clinically useful marker of central nervous system abnormality in Fetal Alcohol Spectrum Disorders (FASD) than common measures of productivity or grammatical complexity and (b) whether combining the rate of grammatical errors with the rate of cohesive referencing errors can improve utility of a standardized narrative assessment task for FASD diagnosis. Method: The method used was retrospective analysis of narrative and clinical data from 138 children (aged 7-12 years; 69 with FASD, 69 typically developing). Narrative analysis was conducted blind to diagnosis. Measures of grammatical error, productivity and complexity, and cohesion were used independently and in combination to predict whether a story was told by a child with an FASD diagnosis. Results: Elevated grammatical error rates were more common in children with FASD, and this difference facilitated a more accurate prediction of FASD status than measures of productivity and grammatical complexity and, when combined with an accounting of cohesive referencing errors, significantly improved sensitivity to FASD over standard practice. Conclusion: Grammatical error rates during a narrative are a viable behavioral marker of the kinds of central nervous system abnormality associated with prenatal alcohol exposure, having significant potential to contribute to the FASD diagnostic process.