Structural plasticity of dendritic spines has been associated with perceptual and motor learning in the mouse neocortex in vivo. In the motor learning tasks, the extent of structural synaptic plasticity was shown to correlate with the improvement in behavioral performance. However, these studies did not unequivocally demonstrate that the neurons displaying the changes were causally related to the improvement of behavioral performance. Last year, for the first time, a study probing the causality of spine formation upon learning was tested. The elimination of newly formed and potentiated spines drove forgetting of a learned behavior. Along this same line of causality, we measured presynaptic structural changes in neurons that are necessary for a perceptual task. We trained adult mice to report the optical microstimulation of a few hundred somatosensory supragranular pyramidal neurons expressing Channelrhodopsin2 (ChR2). We used small head-mounted LEDs to stimulate ChR2-expressing neurons through a permanent cranial window. In daily training sessions head-fixed mice learned to correctly report the presence or absence of the microstimulation in order to obtain a water reward in a “Yes/No” task. On average mice that learned the task reached ≈ 65% accuracy levels within 4 weeks, but some failed to exceed chance level (50%), even after 8 weeks of daily training. Only the axons of the microstimulated neurons could provide the information required for solving this perceptual task. We hypothesize that structural changes of those axons will favor the downstream signaling needed for the behavioral performance. Using two-photon laser scanning microscopy, we repeatedly imaged large volumes1 in the supragranular layers during the baseline and learning periods in order to capture axonal structures of a subset of the ChR2-expressing neurons (which were colabeled with GFP). We developed methods for identifying and tracking large populations of GFP-labeled axonal boutons over time. This enabled the quantification of structural changes in thousands of boutons over months. Statistical modeling was used to analyze bouton dynamics and to characterize synapses that were associated with the improvement of behavioral performance. The results were compared between learners and non-learners in the attempt to identify anatomical correlates of perceptual learning in mouse barrel cortex. Our analysis suggests that animals that learned the task destabilized a higher fraction of boutons present before the training, as compared to controls that did not learn the task but similarly received microstimulation. On top of that, the accumulation of new boutons in learners was above that of non-learners and preceded the improvement in behavioral performance. We did not observe changes on the rate of bouton formation or elimination over time. We conclude that a higher density of new structural synapses were present on the final subassembly configuration on learners and these new synapses might be involved in the learning of the new perceptual task.