Close to 20% of secondary school students in Amsterdam - and elsewhere - transfer between secondary schools at some point, even when initially placed in their most-preferred school. School switching is costly for the students involved and disrupts the learning environment of their former and new classmates. Using data from the Amsterdam secondary-school match linked to administrative registers, we show that switching can be predicted by hard-to-rationalize initial school choices. Over 60% of switchers can be correctly identified at the admission stage. Simulations indicate that encouraging predicted switchers to adjust their preference ranking of schools could reduce the switching rate by almost 15%.