Metamemory monitoring is based on prior beliefs about memory and/or processing experience during study (Koriat, 1997). Therefore, the relative contributions of belief and experience should explain how metamemory cues affect different metamemory judgments. The authors modelled participants’ Bayesian inferences in metamemory monitoring (Hu et al., 2021) to compare the relative contributions of belief and experience to metamemory about item and source memory in source monitoring. The authors reanalyzed data from 17 participant groups in seven published studies (total N = 1100) that tested effects of schema-based expectations on metamemory and source monitoring. All participants studied expected and unexpected source–item pairs (e.g., “toothbrush in the bathroom” vs. “microwave in the bathroom”) and made trial-wise metamemory judgments about item memory (Judgments of Learning, JOLs) and/or about source memory (Judgments of Source, JOSs). Later, actual item and source memory were assessed. As previously documented, participants showed expectancy illusions on JOLs and JOSs, that is, they predicted better item and source memory for items from an expected source, but actual item memory was unaffected by expectations and source memory was better for items from an unexpected source. The Bayesian inference model showed that belief and experience jointly contributed to metamemory judgments. In expected trials, belief and experience contributed equally to metamemory. In unexpected trials, belief contributed more than experience. Critically, the contributions of belief and experience explained differences between JOLs and JOSs. JOSs showed a stronger expectancy illusion and a stronger contribution of experience than JOLs. Reliance on belief was associated with weaker illusions.