The neural mechanisms of visual perceptual learning (VPL) remain unclear. Previously we found that activation in the primary visual cortex (V1) increased in the early encoding phase of training, but returned to baseline levels in the later retention phase. To examine neural changes during the retention phase, we measured structural and functional connectivity changes using MRI. After weeks of training on a texture discrimination task, the fractional anisotropy of the inferior longitudinal fasciculus, a major tract connecting visual and anterior areas, was increased, as well as the functional connectivity between V1 and anterior regions mediated by the ILF. These changes were strongly correlated with behavioral performance improvements. These results suggest a two-phase model of VPL in which localized functional changes in V1 in the encoding phase of training are followed by changes in both structural and functional connectivity in ventral visual processing, perhaps leading to the long-term stabilization of VPL.Visual perceptual learning (VPL) is defined as long-term visual performance improvements after visual experiences, and is thought to reflect brain plasticity in adults [1][2][3] . However, the underlying neural mechanisms of VPL are not completely understood 4 . The brain area(s) that change in association with VPL remain hotly debated in the field. Some researchers have suggested that VPL takes place in the early visual cortex, citing that VPL is specific to the trained feature/location. For example, VPL of a texture discrimination task is specific to the trained visual field quadrant and does not transfer to other quadrants 5 . Such location specificity suggests the involvement of the early visual cortex, in which visual processing occurs locally [5][6][7] . A number of human functional magnetic resonance imaging (fMRI) studies have documented the involvement of the primary visual cortex (V1) [8][9][10][11] in VPL. On the other hand, the majority of animal studies have failed to find evidence of the involvement of V1 neurons, in contrast to results from human studies [12][13][14] (but see refs [15][16][17][18] ). The results of a recent study suggest a reason for the discrepancy 7 , namely, a difference in time course of learning development. Yotsumoto et al. 7 showed that in the early phase of VPL training, performance increases are accompanied by activation enhancement in the region of V1 corresponding to the trained stimulus. However, in the later phase of VPL training, after performance increases have reached a plateau, V1 activation returns to baseline levels, even while performance on the trained task remains high. These results are in accord with a two-phase model of VPL, in which V1 is involved in the encoding of VPL, and in a later phase different regions and/or aspects of the brain are involved in the long-term retention of VPL. This model accounts for both the involvement of V1 in human studies and the absence of V1 activation changes in monkey single-unit recording studies, and may effectiv...