Experimental and subjective evidence shows that human perception is highly ambiguous, primarily since sensory signals often carry too little information to disambiguate different states of the world. Several theoretical accounts have provided suggestions how this ambiguity may be resolved. Theories of predictive coding suggest that nervous systems continuously generate predictions about upcoming states of the world and perception results from a comparative process of predictions and real-world sensory input. Consistent with this, the Bayesian brain hypothesis states that nervous systems encode both prior expectations and stimulus information probabilistically and that subjective percepts reflect posterior beliefs resulting from a Bayesian inference process. However, robust evidence for these hypotheses come from a limited range of experimental paradigms, often failing to provide conclusive answers and the neuronal mechanisms underlying these processes remain unclear. I will present new evidence from studies using a combination of psychophysics, magnetoencephalography and transcranial current stimulation that shed more light on how the brain integrates expectations with sensory signals.