Modern neuroscience tends to regard the brain as a predictor of the external environment, which can actively infer forthcoming sensory information based on both the prior knowledge and current sensory inputs, interacting with the external environment effectively. The predictive coding stands at the center of theories describing the mechanism of prediction. Reviewing the studies about the predictive coding, especially the ones in visual and auditory modality, can offer a new theoretical perspective for a more in-depth comprehension of the brain’s operational mode. This article first described the content of the predictive coding. And then it reviewed typical studies on the interactions between the predictions and sensory inputs, in terms of common paradigms, typical phenomena, and controversies. The third part introduced the endogenous neural representations of prediction, which are stimulus-independent, mainly from two aspects, i.e., the neural representations of prediction when the stimulus was omission, and prediction-related neural oscillation patterns. Furthermore, this review presented major neurophysiological evidence and neural structures that support the hierarchical structure. Finally, the development of the predictive coding related research is prospected from the aspects of deepening theoretical research, helping disease diagnosis and treatment, and inspiring the brain-computer interface technology. It is believed that a deeper comprehension of the computational model and neural representations of prediction coding in visual and auditory neural activity would offer up new perspectives on how perceptual neural activity in the brain functions.