Improper Motions in Lensed QSOs

We argue that individual images in multiple-image QSOs could easily have substructure at the level of $0.1''$ (i.e., unresolvable even with HST); microlensing within such substructure would cause centroid shifts, observable even from the ground as pseudo proper motions. We present a model of the four-image system 2237+0305 in which Image B shows such ``improper motions" of order of $0.01''$ over a few years.