This paper implements the traditional initiation-feedback-response (IRF) model in an intelligent English grammar learning assistant system for junior high school students. The system, named the IRF-based English Grammar Learning Assistant (IRF-EGLA) System, is constructed to deliver automatically initiation questions and, based on the learners' responses, give proper feedbacks in an e-learning environment according to the IRF dialogue strategies. This study uses a quasi-experiment with pre-tests and post-tests to examine the learning performance of the participants and a questionnaire to examine participants' perceptions of the usability of the system. In the experiments, 100 eighth-grade students were assigned to three groups according to their English classes. In the experimental group, 35 participants used the IRF-EGLA system. In the control group A, 31 participants used a non-IRF EGAL system. In control group B, 34 participants did not use any EGAL system, but classroom lectures. The experimental results indicate that the IRF-EGLA system is more effective than the non-IRF EGAL system with respect to learning performance, but is no better than classroom lectures. However, after excluding the underachievers, who, from their learning profiles, have poor English vocabulary capacity and lose their learning interest very soon, the experimental group does perform better than the other two groups. From the results of the questionnaire, most of the participants believe that the IRF-EGLA system could help them learn English grammar rules effectively and, in the meantime, increase their interest in learning English.