While there has been a proliferation of research on assessment for learning (AFL) over the past two decades, L2 writing assessment has tended to focus much more on assessment of learning (AOL) than AFL. This study seeks to investigate the implementation of AFL for EFL writing within an examination-driven AOL system in Hong Kong, its possible impact on students' motivation and writing performance, as well as the factors that might facilitate or inhibit its uptake. Multiple sources of data were collected, including questionnaires, interviews, pre- and post-tests, and lesson observations. The findings showed that while teachers strengthened planning and pre-assessment instruction, sharing learning goals with students and using feedback forms to provide feedback and align assessment with instruction, they had to adhere to conventional practices that required detailed attention to errors and summative scores, and were unable to engage students in multiple drafting and peer evaluation on a regular basis. Results about students' motivation are mixed, suggesting some tension between innovative and traditional assessment practices. Pre- and post-tests indicated some improvement in students' writing performance, with teachers believing that AFL did have a role to play. The factors that facilitated and constrained teachers' attempts to implement assessment for learning are discussed, and implications drawn. (Contains 2 tables.)