In this issue of Neuron, Konovalov and Krajbich (2018) argue that a Bayesian inference is employed when learning new sequences and identify distinct brain networks that track the uncertainty of both the current state and the underlying pattern structure.