Interoception, which means sensations of inner body signals and regulation of the body to maintain homeostasis, has been considered to play key roles in emotion and decision-making. Recently, the theory of predictive processing has been focused as a potential theoretical framework to explain the association between interoception, emotion, and decision-making. In this perspective, interoception is considered as not just a bottom-up processing but to emerge through comparison between predictions by inner models in the brain and actual signals (prediction error). The brain regulates physiological systems by minimizing prediction errors. Success to reduce prediction error might be linked with positive emotions and expansion and maintenance of prediction error might result in negative emotions. Furthermore, success of prediction error reduction and accompanying positive emotions can work as reward and increase values of related behaviors and objects. Inversely, failure of prediction error reduction devaluates related behaviors and objects. Consequently, probabilities of choices of the behaviors and objects can be modulated. Such a theoretical framework can consistently combine several important psychological phenomena such as interoception, emotion, reward, and decision-making. Computational models describing dynamics of the predictive processing of interoception can explain empirical data in neuroimaging and psychophysiological experiments. In this theoretical framework, emotional intelligence can be considered as abilities to efficiently reduce interoceptive prediction errors and to appropriately regulate bodily and emotional states based on interoception. This framework should shed a new light for understanding about human nature, and furthermore for developing of new ways to improve emotional intelligence.