The notion that unconscious Bayesian inference underlies perception is gaining ground. Predictive coding approaches posit that the state of the world is inferred by integrating, at each level of the perceptual hierarchy, top-down prior beliefs about sensory causes and bottom-up prediction errors. In this framework, percepts correspond to a top-down stream of beliefs that best 'explain away' sensory signals. Although such frameworks are gathering empirical support, subjective facets of perception remain unexplained from these perspectives. This thesis combines behavioural, computational and neuroimaging methods to examine how subjective visual confidence can be accounted for in a predictive coding framework. Experiment one shows that, behaviourally, perceptual expectations about target presence or absence both liberalise confidence thresholds and increase metacognitive accuracy. These results are modelled in a signal detectiontheoretic framework as low-level priors shifting the posterior odds of being correct. Using EEG, experiment two reveals that influence of expectations on decision and confidence oscillates with the phase of pre-stimulus alpha oscillations. This means that prior to target onset, both objective and subjective decisions have been rhythmically biased by the periodic recruitment of expectations to visual areas. Using fMRI, experiment three shows that in the post-stimulus period, expectations and sensory signals are integrated into confidence judgements in right inferior frontal gyrus (rIFG). Furthermore, this process recruits orbitofrontal cortex and bilateral frontal pole, which represent top-down influences, and occipital lobe, which represents bottom-up signals. Together, these results suggest that expectations shape subjective confidence by biasing the posterior probability of the perceptual belief.