For decades, policymakers and researchers have used value-added models that rely solely on student test scores to measure teacher quality. However, since teaching ability is multi-dimensional, test-score value-added measures of teacher quality may not fully capture the impact of teachers on students. In this paper, we use test-score and non-test-score measures of student achievement and behavior from over a million students in the Los Angeles Unified School District to estimate multiple dimensions of teacher quality. We find that test-score and non- test-score measures of teacher quality are only weakly correlated, and that both measures of teacher quality affect students’ performance in high school. A teacher-removal policy simulation that uses both dimensions of teacher quality improves most long-term student outcomes by over 50 percent compared to a policy that uses test scores alone. Our results also show that the long-term effects of teachers in later grades are larger than in earlier grades, that performance in core elementary school subjects matters more for long-term outcomes than other subjects, and that non-test-score dimensions of ability capture additional sources of peer effects.