High-stakes tests have been employed widely to engineer curriculum innovation, or achieve intended washback in education. But our understanding of the role of high-stakes tests as an agent for change is limited due to the small number of empirical studies available on this issue. This paper reports on a washback study which focuses on the writing task in the National Matriculation English Test (NMET) in China and describes both its actual and its anticipated effects on secondary school teaching. Participants in the study consisted of an extensive sample of the NMET stake-holders, comprising test constructors, teachers, and students. Data collection instruments included interviews, classroom observations, and questionnaires. The results show that although writing was practised in schools, it was not taught in the way intended by the test constructors. The communicative features of the NMET writing task were not observed in the school practice that prepared students for this task. Both teachers and learners neglected the communicative context of writing while emphasizing the testing situation and the assumed preferences of the markers. This urge to raise scores in a real test situation suggests that high-stakes tests are not an efficient agent for change despite their crucial role in education. (Contains 6 tables and 13 notes.)