In the US, many federal, state and local school improvement policies rely on teacher professional development (PD) to foster classroom change. Past research suggests PD that has a content focus is the most effective, but that even content-focused PD varies in its effectiveness. Through in-depth interviews of teachers participating in a middle school science PD randomized control trial in the US, we find that what teachers learn in PD varies significantly based on their prior knowledge and experience. This paper explores several hypotheses about how content knowledge and teacher learning interact. We conclude that the next step toward improving teacher PD is to calibrate learning opportunities to teachers' prior knowledge.