The study developed a model of linguistic constructs to predict writing quality for college basic writers and analyzed how those constructs changed following instruction. Analysis used a corpus of argumentative essays from a quasi-experimental, instructional study with 252 students (MacArthur, Philippakos, & Ianetta, 2015) that found large effects (ES = 1.22) on quality of argumentative writing. Coh-Metrix (McNamara, Graesser, McCarthy, & Cai, 2014) was used to analyze the essays for lexical and syntactic complexity and cohesion. Structural equation modeling found that referential cohesion (p < 0.001) and lexical complexity (p < 0.01) positively predicted quality on posttest essays while syntactic complexity (p < 0.001) was negatively related to quality. Length explained 30% of variance in quality; the full model explained 48.7%. Confirmatory factor analysis was used to impute factor scores for pretest and posttest essays. Analysis of covariance using these factors found that the treatment group wrote posttest essays with greater lexical complexity (p < 0.01) and referential cohesion (p < 0.01) and less use of connectives (p < 0.05) than a business-as-usual control group. [This content was accepted for publication in "Reading and Writing: An International Journal" in April 2018.]