<p>Usage-based theories have influenced English grammar instruction by focusing on real-world language use. Yet, grammar teaching remains relatively unchanged by research findings. This study, therefore, investigated the effect of corpus based instruction on students’ grammatical accuracy in writing skills. Employing a pre-test post-test quasi-experimental design, participants were assigned into an experimental group learnt through corpus based instruction and a control group received traditional instruction. The investigation focused on modal auxiliary verbs of ability identified as problematic areas. Data was collected through written tests and analyzed quantitatively. The Ethiopian Students Corpus and British National Corpus were compared for frequency distribution and error patterns of target grammar feature. Pre-test and post-test results showed significant improvements in both groups with the experimental group outperforming the control group. Regarding to frequency distribution and error patterns, the results showed that "can" was more common in the Ethiopian Students Corpus, while "could" was frequent in the BNC, with no use of "was/were able to" in the Ethiopian corpus. Overgeneralization and subject-verb agreement errors were reduced in the experimental group. From the results, it was recommended that authentic linguistic data be integrated into grammar lessons could actively engage students with language forms and meanings</p>