In the study of metamemory monitoring, it is common to make assumptions about how predictions of future memory performance (metamemory judgments) relate to latent memory processes. In practice, however, most research in this area focuses on empirical measures of memory performance rather than directly linking metamemory judgments to underlying latent processes. One modeling approach that allows for the measurement of latent (memory) processes is multinomial processing tree (MPT) modeling. However, linking continuous metamemory judgments to latent MPT parameters is not straightforward. Previously, this has been achieved by Vincentizing (binning) continuous judgments and relating them to aggregate-level MPT parameters. While this represents a significant step forward, it results in a substantial loss of information and prevents the analysis of individual differences in these relations. Therefore, we propose a novel approach using Bayesian hierarchical MPT models in which continuous metamemory judgments are directly included as trial-level predictors of person-wise latent model parameters. We compare this approach to the Vincentization method in a simulation study and by applying it to published data on the relationship between judgments of learning (JOLs) and latent familiarity- and recollection-based memory processes.