We used multilevel analysis to examine the predictive validity of scores from the Framework for Teaching (FFT), the observation measure used most often to evaluate teachers' instruction. We investigated how well 81 kindergarten teachers' FFT scores for eight reading and eight mathematics lessons observed throughout the year predicted students' year-end achievement and motivation in reading and mathematics, controlling for students' sex, ethnicity, and achievement entering kindergarten. Standardized reading and mathematics achievement were each predicted by FFT scores; however, they accounted for very little of the overall variance in students' achievement: 2.5% for reading and 1.3% for mathematics. Neither students' end-of-year criterion-referenced achievement nor motivation were predicted by FFT scores.