We usually think that as we emerge from childhood, our brains become less plastic, making learning effortful and highly specific. Recent findings however challenge this view, suggesting that even adult perceptual learning, often considered the most specific form of learning, has the potential to generalize across training conditions. This questions classical theories positing that perceptual learning changes encoding in early sensory areas, as the functional properties of these areas cannot account for generalization. Building on recent computational models, I propose instead that PL relates to decoding, that is, how information from sensory areas is communicated and read out by higher areas to make decisions. Decoding accounts are theoretically attractive yet technically challenging to test, as they require a multiscale brain investigation, i.e. tracking perceptual learning across networks, areas, and single neurons. In this project, we address these challenges by combining functional magnetic resonance imaging and electrophysiological recordings during learning tasks. This allows us to test where perceptual learning takes place in the brain (in sensory areas and/or in higher-order brain areas), and what computations the neurons in these areas perform. This project, at the intersection of neuroscience, psychology and computational theory, will set forth the foundations for a mechanistic investigation of perceptual learning at an unprecedented level of detail, bridging multiple scales from whole-brain networks down to single neurons. Ultimately, this innovative framework will help us understand the building blocks of adult brain plasticity, and how to optimize rehabilitation and educational applications.