Annual student testing may make it possible to measure the contributions to student achievement made by individual teachers. But would these "teacher value-added" measures help to improve student achievement? I consider the statistical validity, purposes, and costs of teacher value-added policies. Many of the key assumptions of teacher value added are rejected by empirical evidence. However, the assumption violations may not be severe, and value-added measures still seem to contain useful information. I also compare teacher value-added accountability with three main policy alternatives: teacher credentials, school value-added accountability, and formative uses of test data. I argue that using teacher value-added measures is likely to increase student achievement more efficiently than a teacher credentials-only strategy but may not be the most cost-effective policy overall. Resolving this issue will require a new research and policy agenda that goes beyond analysis of assumptions and statistical properties and focuses on the effects of actual policy alternatives.