The time course of the decay of spatial representations used for planning and controlling manual aiming has not been established. The authors' purpose in the present investigation was to generate a psychometric function for memory-guided reaching movements. Eight university-aged students performed a reciprocal tapping task for 10 s. Participants could see the targets for 5 s; then, vision of the targets was occluded. The present findings provide mixed support for 2 prominent theories concerning memory representations. Variability increased concurrently with the removal of vision of the targets, supporting the real-time hypothesis. However, a brief plateau in the curve was apparent for approximately 2 s after vision was removed, consistent with the use of a highly accurate representation for action.