If motion that one has been looking at for some time suddenly stops, or if one shifts one's gaze to a static object, one will see motion in the opposite direction: the motion after-effect. If two transparent surfaces move with different speeds in different directions, then the direction of the motion after-effect will depend on the test pattern. For such transparent surfaces both the local motion and the global percept have two components. When looking at a normal moving object, there is only one perceived global motion. However, we know that locally there can be considerable ambiguity (the aperture problem). Does one adapt to all the local components, including those that one does not perceive, or only to the perceived global motion? We designed a stimulus that is perceived to be a fast rotating object, but also has a slow local radial component of motion. By selecting an appropriate test pattern we could either get a radial or a rotating motion after-effect. Thus we show that adaptation to motion must (also) occur at a stage at which local motions have not yet been integrated to give a unified percept.