Both ambiguous inference from current input and internal belief from prior input causes uncertainty. The uncertainty is typically manifested as a normal distribution at behavioral level when only current inference is manipulated as variable. When prior belief is varying, some decision relevant neural representations are dissociated. Under this circumstance, it is unclear how to describe the uncertainty and how dissociated neural representations cooperate to control the uncertainty. By simulating an unpredictable environment, which incurs conflicting valence-dependent prior beliefs, we found that a behavioral outcome, waiting time, does not follow a normal, but a log-normal distribution. By combining electrophysiological recordings, computational modeling, optogenetic manipulation, scRNA-seq and MERFISH, we showed that the formation of this behavioral outcome requires the temporally hierarchical cooperation of the neural representation of decision confidence and B230216N24Rik marked neural representation of positive and negative belief in the medial prefrontal cortex (mPFC). In summary, our results provide a mechanistic link between the dynamics of valence-dependent prior beliefs and behavioral uncertainty.