This paper uses data from the American Mathematics Competitions to examine the rates at which different high schools produce high-achieving math students. There are large differences in the frequency with which students from seemingly similar schools reach high achievement levels. The distribution of unexplained school effects includes a thick tail of schools that produce many more high-achieving students than is typical. Several additional analyses suggest that the differences are not primarily due to unobserved differences in student characteristics. The differences are persistent across time, suggesting that differences in the effectiveness of educational programs are not primarily due to direct peer effects.