Binocular saccades in response to briefly flashed, memorized disparate targets (different for the two eyes) become disconjugate following repeated trials. After 15 min of such training, the disconjugacy persists, even when the target to memorize is no longer disparate. This study examines the hypothesis that disparity memorization has a motor basis. We report here three experiments in which subjects were trained for 15-min periods. In experiment 1, subjects made no saccade after target presentation (static training); in experiment 2 subjects intended to make a saccade, but they actually made a saccade in only 10% or 20% of the trials; in experiment 3 subjects made anti-saccades. For all three experiments, the flashed target was disparate and the memory delay for each trial was 1 s. To examine the effects of learning for all three experiments, before and after training, we recorded memory-guided saccades to non-disparate targets (monocular viewing). Experiments 1 and 2 produced inconsistent (before/after training) changes in the disconjugacy of saccades. Thus, the disparity of potential saccade targets had no lasting effect on the disconjugacy of saccades if a saccade was not made. In contrast, the anti-saccades in experiment 3 developed a disconjugacy opposite to the disparity of the remembered target. These findings indicate that the execution of the saccade is necessary to memorize disparity of the target.