Formulaic language has traditionally not been given sufficient consideration by ELT materials developers. Thus, the lexical bundles (LBs) used in reading texts found in high stakes language tests may not reflect the kinds of phrasings used in the target domain of the test calling its validity into question. To investigate this issue in the English section of the Korean college scholastic ability test (CSAT), all of the reading texts from the 2017 through 2021 CSAT examinations were used to build a CSAT corpus which was compared with the Manually Annotated Sub-Corpus (MASC) reference corpus and Martinez & Schmitt’s (2012) PHRASE list. LBs were extracted for the analysis using Wordsmith 8. The analysis involved examining the frequency and distribution of lexical bundles and other linguistic features in the two corpora. It was found that only two LBs were shared between the top ten most frequent phrases in the CSAT and MASC, indicating that most of the high-frequency LBs in the CSAT were not the same as those in the MASC. The structures and functions of LBs were also compared, and while there were some minor differences between the two corpora, they were generally similar. Additionally, the CSAT contained fewer useful LBs than the MASC, but the difference was not large. Based on these findings, it is suggested that CSAT creators give more consideration to the formulaic language used in the adapted texts used in the CSAT.