Abstract: The cerebellum is a key brain region involved in associative learning, and in particular for generating predictive sensorimotor associations. To mediate such learning, convergent input from two main pathways is thought to be required; the climbing fiber and granule cell pathways. Using multiphoton imaging in awake behaving mice, we have investigated how each of these pathways encodes the sensory and motor information necessary for learning. These studies have revealed surprising results that extend current views of cerebellar learning. Specifically, we have found that cerebellar climbing fibers can exhibit reward-related responses that are consistent with many of the predictions of reinforcement learning, in contrast with the long-held view that the cerebellum operates exclusively according to supervised learning principles. In addition, we have found that granule cells generate sparse population codes that rely on local synaptic inhibition to enable pattern separation and learned sensorimotor discriminations. I will discuss the implications of these results in the context of cerebellar associative learning.