For eye movements made to visual targets, the brain must transform the retinotopic coordinate frame of the visual system to that of the oculomotor plant. Ideally, responses should exactly match target demands. However, during eye movements to remembered targets, responses are spatially distorted. The transformation does not retain accurate retinotopic registration, having both constant and variable components of error. Generally, the constant pattern of distortion appears as a hypermetria for upward saccades and a hypometria for downward movements. Most of the error accumulates during the first 800 msec of memory-contingent delay. The results are interpreted with respect to theories of how spatial information may be coded and transformed.