This study aimed to explore the relationship between students' understanding of the National Matriculation English Test (NMET) and their learning practices through standard multiple regression (SMR) and structural equation modeling (SEM) with the purpose of unraveling the working mechanism of washback. A total number of 3105 Chinese senior high school students participated in the questionnaire survey. Data were processed and analyzed using SPSS 21.0 software to conduct descriptive statistics, exploratory factor analysis, and standard multiple regression and using AMOS 21.0 software to perform confirmatory factor analysis (CFA) and SEM. The findings revealed that students demonstrated a better understanding of test designs than test demands, and that their understanding of the test significantly predicted their learning practices. Specifically, students' understanding of both test designs and test demands significantly predicted teaching-related learning and test preparation. However, understanding of test demands was found to be a stronger predictor of teaching-related learning and test preparation than understanding of test designs and was the only predictor of communicative learning. These findings have significant implications for students, teachers, and educational authorities, highlighting the need for targeted support and guidance to enhance students' understanding of high-stakes testing and its impact on learning behaviors.