In human subjects, we investigated the accuracy of goal-directed arm movements performed without sight of the arm; errors of target localization and of motor control thus remained uncorrected by visual feedback, and became manifest as pointing errors. Target position was provided either as retinal eccentricity or as eye position. By comparing the results to those obtained previously with combined retinal plus extraretinal position cues, the relative contribution of the two signals towards visual localization could be studied. When target position was provided by retinal signals, pointing responses revealed an over-estimation of retinal eccentricity which was of similar size for all eccentricities tested, and was independent of gaze direction. These findings were interpreted as a magnification effect of perifoveal retinal areas. When target position was provided as eye position, pointing was characterized by a substantial inter-, and intra-subject variability, suggesting that the accuracy of localization by extraretinal signals is rather limited. In light of these two qualitatively different deficits, possible mechanisms are discussed how the two signals may interact towards a more veridical visual localization.