While a large body of research has been accumulated that provides reliability and validity evidence for L2 integrated writing tasks, relatively little research has been conducted to examine integrated writing tasks as a means to provide diagnostic insights for teachers and learners. The current study aims to fill in this lacuna by applying a log-linear cognitive diagnostic model (LCDM) to reading-to-write integrated writing data collected from 315 Chinese college-level English as a Foreign Language (EFL) examinees. For this study, the integrated writing task was conceptualized as consisting of language use, source use, and content, with each of these unobservable attributes measured by surrogate indicators. Results showed that all the pairs of postulated attributes were positively correlated. However, the association between language use and content (r = 0.36) was not as strong as that of either language use and source use (r = 0.74) or source use and content (r = 0.90). Also, item parameters indicated that language use is more important than other attributes for obtaining a passing score for writing features. Lastly, the test-taker classification showed that it is impossible to master source use without other attributes, demonstrating the dependence of source use on other attributes. Implications for teaching are discussed.