Humans typically perceive their visual world as stable and continuous, despite frequent shifts of the retinotopic reference frame caused by saccades. This visual stability is paralleled by the curious case of afterimage movement across saccades: Albeit being stabilized retinotopically, afterimages appear to move in egocentric space wherever the eye moves. While this phenomenon has been known for centuries, the exact relationship between eye and afterimage position remains elusive. Using custom-built presentation hardware in complete darkness, we subjected twelve human observers to a novel task, in which they viewed a bright foveal inducer pulse during fixation, executed a saccade to a previously cued location, and performed spatial localization of the emerging afterimage with respect to a sequence of briefly flashed probes. This psychophysical tracking of afterimages was accompanied by high-precision eye tracking, allowing us to estimate afterimage-movement gain by fitting a dedicated computational model to the data. We found that the size of afterimage movement was tightly contingent upon the size of the corresponding eye movement, but yielding significantly hypometric gains which averaged to 0.94 and were remarkably stable across experimental sessions. Gains were unaffected by (even false) post-saccadic visual feedback and remained constant when saccade amplitudes were successfully adapted to significantly lower gains. Regardless of saccadic adaptation, afterimage-movement gains were inversely related to saccade gain - a finding that could be explained by a parsimonious efference-based model of head-centered visual localization. Our results suggest that afterimage movement across saccades is driven by feedforward prediction of the visual consequences of impending saccades and demonstrate the potential of the afterimage-tracking technique for studying sensorimotor processes involved in active vision.