Over the past decade, the New Zealand government has created a set of resources to support teachers’ use of assessment for learning. These include Assessment Tools for Teaching and Learning (asTTle), a software program enabling teachers to create personalized but standardized tests for diagnostic purposes, and Assess to Learn (AtoL), an intensive professional development program. These resources were expected to increase teacher agreement that improvement is the major purpose of assessment. Instead, a 2008 sample of teachers completing the Teacher Conceptions of Assessment questionnaire showed significantly higher agreement that assessment is about school accountability than participants in previous national surveys. Unlike previous surveys, the correlation between school accountability and improvement conceptions was not statistically significant. However, as only the improvement conception predicted the practices teachers used to define assessment (β = .32), it appears that these teachers still saw many of the practices they used in the classroom (e.g. oral and interactive assessments) as improvement-oriented. Interviews with twenty-six of the surveyed teachers identified that while a few saw the new resources as contributing to improvement and accountability purposes, a larger group failed to make that connection. This second group seemed to be unable to accept that tests, an assessment genre traditionally associated with school and student accountability, could be meaningfully used for improvement at the classroom level. These data show that schools and individuals mediate the implementation of any policy initiative and can therefore cause it to have a range of often unintended consequences. In light of this, the thinking of teachers and other educational stakeholders should be taken into account when enacting policy changes.