Background: This study investigates linguistic characterizations in the form of linguistic co-occurrence patterns in discussion sections of English research articles (RAs) in an engineering discipline (i.e., chemical engineering) and linguistic variations that distinguish discussion sections of high-impact articles from those in low-impact articles. Research questions: 1. What underlying linguistic characterizations are salient in RA discussions in chemical engineering? 2. Are there any differences in the identified linguistic characterizations of discussion sections between high- and low-impact RAs? Literature review: In the process of composing RAs, the discussion section is a difficult and challenging part-genre to write. The rhetorical organization of RA discussions has been examined extensively through Swales's English for Specific Purposes genre analysis. However, the linguistic characterizations of RA discussion sections remain unclear and the question of whether discernible differences exist between discussions of high- and low-impact RAs in a specialized engineering discipline remains unanswered. Methodology: This study used Biber's multidimensional (MD) analysis method. In response to the first research question, factor analysis (in this study, principal component analysis) was adopted to identify the linguistic characterizations in the form of linguistic co-occurrence patterns (“dimensions”) in 213 RA discussion sections extracted from chemical engineering RAs. To answer the second question, the independent t-test was implemented to compare the high- and low-impact RA discussion sections in the identified dimensions. Results and conclusions: Six linguistic characterizations in the form of linguistic co-occurrence patterns were identified in RA discussion sections: 1. involvement and interactivity, 2. non-narration versus narration, 3. evaluative statements with further explanations and elaborations, 4. informational density, 5. stating results/claims, and 6. expression of denial relationships toward statement or experimental findings. The results suggest the linguistic characterizations in RA discussion sections and interesting differences in the high- and low-impact RA discussion sections, especially in Dimensions 1, 3, and 5. Reasons for the linguistic variations in the identified dimensions are discussed, followed by the pedagogical implications for reading or writing RAs for international scientific communication.