Diagnostic language assessment (DLA) consists of three core components: diagnosis, feedback, and remedial learning, yet the effectiveness of remediation is largely ignored. Utilizing the online platform of Udig, this study examined the effects of the computer-mediated DLA on the grammar learning of Chinese EFL learners with high and low proficiency level. Results showed that the computer-mediated DLA system was more effective in improving learners’ grammar learning than traditional approaches. Compared with the control groups, the experimental groups made greater improvement in three sub-skills, namely, compound sentence, inversion, and modal verb. The specific improved sub-skills varied between high and low level EFL learners, with greater improvement in subjunctive mood and inversion for high-level students, but more improvement in complex sentences for low level students. These respective sub skills correspond closely to the relative weaknesses for both levels of students, indicating that DLA might work more effectively on learners’ weak skills. Results also showed that, compared with low-level students, high-level students improved significantly in the preposition sub-skill, suggesting that EFL proficiency might be a factor impacting DLA effects on grammar learning.