We use a novel database of the preservice apprenticeships ("student teaching placements") of teachers in Washington State to investigate the relationship between mentor effectiveness (as measured by value added) and the future effectiveness of their mentees. We find a strong, positive relationship between the effectiveness of a teacher's mentor and their own effectiveness in math and a more modest relationship in English Language Arts. The relationship in math is strongest early in a teacher's career, and would be positive and statistically significant even in the presence of non-random sorting on unobservables of the same magnitude as the sorting on observables. This suggests that at least some of this relationship reflects a causal relationship between mentor effectiveness and the future effectiveness of their mentees in math.