As reflected in literature, many researchers suppose that instead of treating vocabulary as single items, a stronger emphasis should be placed on formulaic language (e.g., lexical bundles (LBs)) due to its effectiveness in academic writing (Boers, 2020; Conrad & Biber, 2005). However, there is a lack of studies that have directly analyzed the LBs used by L2 learners in Vietnam and their potential impacts towards students’ performances. To bridge this gap, the study aims 1) to identify the most frequently used LBs in academic texts by Vietnamese undergraduates, distinguishing in two small datasets, namely high-scoring group corpus (13,737 words) and low-scoring group corpus (13527 words), 2) to compare the frequency of these LBs in each group aligned with target bundles - the Academic Formular List (AFL) proposed by (Simpson-Vlach & Ellis, 2010), and 3) to investigate the potential elements which may contribute to the differences in students’ writing performances by using Biber et al.’s taxonomy (2004). The findings reveal that students in the high-scoring group use a greater variety of lexical bundle types compared to their low-scoring counterparts. However, when comparing the use of LBs to the AFL, the low-scoring group demonstrates a higher degree of similarity, deploying 17% of the targeted bundles, while the high-scoring group uses 11%. This suggests that higher-scoring students exhibit a greater ability to use more unique multi-word phrases rather than relying mainly on frequently used bundles. Regarding to the writing quality, a comprehensive analysis of discourse functions of LBs is conducted, indicating numerous differences between the two groups. Particularly, students achieving higher scores deploy more referential bundles and discourse organizers, which increases both coherence and cohesion of their texts. This observation suggests a potential impact on the disparities in student scores. Key words: lexical bundles, high-scoring, lo