Distinguishing between perception and thought is a vacuous task. At least this is what most adopters of predictive coding accounts express. Here I want to argue for the opposite. Although I concur that perception can no longer be equated with strictly bottom-up processing, I argue that thought, in virtue of being at the top of the hierarchy, can be equated with a distinctive kind of process: It predicts but is not predicted by any other level. Using this argument and some recent collaborative experimental work on the much discussed example of racial biases in vision, I show why it makes a difference to the way we frame the issue of whether thought influences perception : What we have is a much more tractable and interesting problem of how much cognitive and metacognitive control we have over our perceptual biases.