The present study examined the relative effects of discourse-based and sentence-level grammar instruction on the learning of English present perfect in academic writing, using a sequential explanatory mixed-methods design. Participants were 37 multilingual graduate students enrolled in different sections of an ESL-writing course at a U.S. university. Prior to two 35-minute instruction sessions, the participants completed a pretest composed of acceptability-judgment, forced-choice, and fill-in-the-blank tasks. The same tasks, with different but comparable items, were administered immediately after and four weeks after the completion of the instructions. Retrospective verbalizations were elicited after completion of the delayed post-test to elicit the learners' awareness of the target form and other related tenses in academic writing. While results of the acceptability-judgment and forced-choice tasks show non-significant differences among the treatment and control groups on the pre-post-delayed post-tests, the discourse-based, fill-in-the-blank task results reveal that the discourse-based instruction group outperformed the other groups immediately after instruction. Results from the retrospective interviews corroborate these findings. Implications for discourse-based grammar instruction are discussed.