Two approaches were adopted to study the pattern of connectivity between the cat visual cortex and lateral geniculate nucleus. Fourteen individual cortico-geniculate axons were labeled and reconstructed after intracellular or extracellular injection of biocytin into regions of known receptive-field position and ocular dominance preference, and the distribution of boutons from multi-axon clusters was mapped in three dimensions and compared with the locations of strategically placed geniculate recordings made in the same tissue. The results show that the feedback has an accurate retinotopic component but that individual axons are both more extensive and more selective than described previously. Area 17 feedback axons terminate primarily in layers A and A1, but the distribution of terminal boutons is strongly biased (3:1 ratio) toward the layer that matches their eye preference. Thus, those driven by the contralateral eye preferentially target layer A, and those driven by the ipsilateral eye target layer A1. Each axon also innervates the perigeniculate nucleus (PGN), but the pattern is otherwise variable, suggesting that there are different axonal classes. The terminal fields of individual axons are much larger than described previously, with a maximum spread of 500–1500 microns. Nevertheless, the projection from a given location in area 17 has a center of maximum terminal density 400–500 microns across, which is in retinotopic correspondence with the aggregate receptive field of the cortical cells of origin. The surrounding zone of relatively sparse boutons, however, must permit corticofugal cells to influence visual processing well beyond the regions over which their own responses summate. It follows that any geniculate cell receives corticofugal input covering an equally extensive area of visual space.