In the antisaccade task, subjects must execute an eye movement away from a visual target. Correctly executing an antisaccade requires inhibiting a prosaccade towards the visual target and programming a movement to the opposite side. This movement could be based on the inversion of the visual vector, corresponding to the distance between the fixation point and the visual target, or the motor vector of the unwanted prosaccade. We dissociated the two vectors by means of saccadic adaptation. Adaptation can be observed when systematic targeting errors are caused by the displacement of the visual target during the saccade. Adaptation progressively modifies saccade amplitude (defined by the motor vector) such that it becomes appropriate to the post-saccadic stimulus position, and thus different from the visual vector of the target. If antisaccade preparation depended on visual vector inversion, then rightward prosaccade adaptation should not transfer to leftward antisaccades (which are based on the same visual vector) but should transfer to rightward antisaccades (which are based on a visual vector inside the adaptation field). If antisaccade preparation depended on motor vector inversion, then rightward prosaccade adaptation should transfer to leftward antisaccades (which are based on the same, adapted motor vector) but should not transfer to rightward antisaccades (which are based on a non-adapted motor vector). The results are in line with the first hypothesis, showing that vector inversion precedes saccadic adaptation and suggesting that antisaccade preparation depends on the inversion of the visual target vector.