This study examines the use of recurrent multi-word sequences (lexical bundles) found in a learner corpus of English argumentative essays written by L1 Korean college students at three different proficiency levels: low, mid and high. After compiling a list of the most frequently occurring four-word bundles in the three sub-corpora of the learner corpus, the study categorises them by structure and identifies which bundles appear in more than one sub-corpus. It then identifies frequent bi-grams embedded in the bundles in each sub-corpus to ascertain how the learners at each proficiency level construct multi-word sequences in context. The findings indicate that more proficient learners favour phrasal bundles, often producing them along with post-modifiers and in longer sentences, and thus approximating norms for academic prose. Lower proficiency learners, however, tend to use more clausal bundles, often including first-person pronouns and employing only a few specific verbs such as want and be, all of which are characteristic of spoken and informal registers. Taken together, these findings reflect L2 development in the use of formulaic language as they describe the uses of lexical bundles that are specific to each proficiency level.