If the purpose of an assessment is diagnosing examinees’ knowledge states to improve their learning, more fine-grained information than the overall level of their ability is necessary. Recent advances in diagnostic assessment triggered the development of cognitive diagnosis models (CDMs), such as the deterministic inputs, noisy “And” gate (DINA) model. Although CDMs for language assessments have been applied to reading or listening test data, a CDM may produce more practical results if the construct to measure has specific and well-defined skill attributes as in a grammar assessment. For this study, a grammar test consisting of 40 multiple-choice items was administered to 3,000 Korean learners of English as a foreign language. From the test items, a Q-matrix, which is an essential tool for CDMs, was constructed based on six grammar skills. As a result, skill profiles were obtained for all examinees. This diagnostic information can be used for tailored instruction. Issues with regard to applications of CDMs to language assessments are also discussed.