One of the issues which attracted the attention of language testers in the 1990s was the impact of high-stakes tests on the classroom: what kind of influence did such tests have on teaching and learning and what could educators do to ensure that this was beneficial rather than harmful? Some progress was made in defining notions such as ‘impact’ and ‘washback’, and a number of studies appeared which analysed the relationship between tests and teachers’ and learners’ attitudes and behaviour. There was a growing awareness of the importance of factors other than test design in determining whether tests would have the impact that was desired. These factors also appear in the literature of educational innovation, and it is to this field that some testers turned for guidance on whether test impact could be predicted or controlled. This paper summarises what language testers have learned about test impact in the last decade and discusses what one model of educational innovation has revealed about how tests interact with other factors in the testing situation. It concludes with a set of recommendations about the steps future test developers might take in order to assess the amount of risk involved in attempting to create change through testing.