We investigate the potential of Tree Substitution Grammars as a source of features for native language detection, the task of inferring an author’s native language from text in a different language. We compare two state of the art methods for Tree Substitution Grammar induction and show that features from both methods outperform previous state of the art results at native language detection.