Metacognitive systematic bias impairs human learning efficiency, which is characterized by the inconsistency between predicted and actual memory performance. However, the underlying mechanism of metacognitive systematic bias remains unclear in existing studies. In this study, we utilized judgments of learning task in human participants to compare the neural mechanism difference in metacognitive systematic bias. Participants encoded words in fMRI sessions that would be tested later. Immediately after encoding each item, participants predicted how likely they would remember it. Multivariate analyses on fMRI data demonstrated that working memory and uncertainty decisions are represented in patterns of neural activity in metacognitive systematic bias. The available information participants used led to overestimated bias and underestimated bias. Effective connectivity analyses further indicate that information about the metacognitive systematic bias is represented in the dorsolateral prefrontal cortex and inferior parietal cortex. Different neural patterns were found underlying overestimated bias and underestimated bias. Specifically, connectivity regions with the dorsolateral prefrontal cortex, anterior cingulate cortex, and supramarginal gyrus form overestimated bias, while less regional connectivity forms underestimated bias. These findings provide a mechanistic account for the construction of metacognitive systematic bias.