Academic performance of disadvantaged students has been stubbornly hard to improve, particularly in urban schools.  The Dallas Independent School District (Dallas ISD) addressed this problem in 2015 with a radical change in the structure of teacher compensation, basing it entirely on classroom effectiveness as determined by an evaluation system incorporating achievement growth, classroom observations and student surveys. We evaluate this institutional change with synthetic control methods that compare math and reading achievement in Dallas ISD schools with achievement for schools in other high-poverty Texas districts. We find large and significant positive effects on math achievement that increase steadily over time. For reading, there is no clear evidence of improvement in Dallas ISD relative to the synthetic control. A mechanism analysis shows that changes in teacher composition account for a majority of the math achievement increase.