Multiple-choice assessments are frequently used for gauging teacher quality. However, research seldom examines whether results from such assessments generalize to practice. To illuminate this issue, we compare teacher performance on a mathematics assessment, during mathematics instruction, and by student performance on a state assessment. Poor performance on the written assessment predicted poor classroom performance; likewise, strong performance on the written assessment predicted strong classroom performance. However, classroom performance of teachers in the middle of the written assessment distribution varied considerably. These results also held true using student outcomes as a criterion, leading to implications for both research and policy. (Contains 3 tables and 4 figures.)