PURPOSE: Letters of recommendation (LORs) are an essential component of residency applications. As grades and Step scores have become pass/fail, programs have fewer objective measures to assess applicants. As a result, even greater emphasis may be given to LORs in residency applications. This study aims to evaluate the role of narrative LORs in Match outcomes. METHODS: From the 2020-21 Match, one hundred first-time applicants from US medical schools were randomly selected. Application characteristics and the narrative LOR text were collected from residency applications. Match outcomes and LOR author demographics were determined from publicly available data. Characteristics of LOR text, such as word count (WC) and word frequency percentages, were calculated using the linguistics software Linguistic Inquiry and Word Count 2015. A bag-of-words classification model designed in Matlab R2022b was used to predict Match outcomes based purely on word frequency variation. The text of 299 LORs was used to train the model, and its accuracy was assessed with 52 LORs. The bag-of-words classification model also identified the words most likely to predict Match outcomes, which were termed “differentiation words.” RESULTS: Of the one hundred applicants included in this study, 69% of applicants matched into a PGY-1 plastic surgery residency position. Although most (57%) applicants were female, only 18.7% of letter writers were female. On average, female applicants had LORs with higher WCs (453 vs. 417, p=0.0497). Additionally, female letter writers wrote higher WC LORs for applicants (475 vs. 429, p=0.045). No differences in WC were observed based on academic rank. For applicants who successfully matched, their LORs averaged a significantly higher WC (463 vs. 375, p<0.0001). The bag-of-words classification model was able to predict outcomes based on the LOR text alone with an accuracy of 91.4%. The word frequency of “differentiation words” was also significantly higher in the LORs of those who matched (2.59% vs. 2.22%, p=0.002). CONCLUSIONS: These results demonstrate that even in the absence of numeric data such as Step 1 scores and publication metrics, the written language in LORs can stratify applicants. This work suggests that highly descriptive and positive adjectives such as “bright” and “excellent” are important keywords for evaluators to include in LORs to indicate applicant strength. Finally, plastic surgery programs may consider using machine learning algorithms to assess LORs, not only to save residency programs time but also to mitigate individual reviewer bias.