The human brain can be understood and organised as a topographic map of cortical areas. Dividing the brain into these distinct subsections has long been an ongoing effort, with a number of often disagreeing attempts to create this map – representative across individuals – being made over time. Both surgical and computational techniques have been broadly utilized in this pursuit, focusing on characterizing these chosen cortical areas based on their structural and functional similarities across individuals. Because the general anatomy is fairly well-understood now, computational methods are favoured; a popular approach taken involves measuring the responses of areas of the cortex according to functional magnetic resonance imaging (fMRI) data. We take a related – but modified – approach in this paper, delineating a model that uses the covariances oof the responses across the cortex rather than the cortical responses themselves. An extension of the existing hierarchical, Bayesian, probabilistic and generative model PrAGMATiC, our mathematical formulation of the model assumes and samples from an underlying Wishart, rather than Gaussian, distribution. This will allow the model to learn parameters to describe the functional covariances of responses at vertices across the cortex. Since direct comparisons of functional values, rather than their covariances, are not readily achieved in resting state fMRI, this formulation will be able to identify cortical parcellations using resting state fMRI data by providing a framework under which comparisons are possible.