Cognition consists of processes operating on representations. These processes are implemented by interdependent transformations of neural responses across brain regions. Modelling these transformations is necessary to understand the neural bases of cognitive processes. In this talk, I will start with a discussion of current methods used to investigate dynamic interactions between brain regions, focusing on Granger Causality and Dynamic Causal Modeling and on their strengths and limitations. I will then introduce a method we have developed to overcome these limitations. In the end, I will integrate these ideas with multi-voxel pattern analysis (MVPA) and representational geometry, reaching a synthesis of these two traditions of fMRI data analysis to study the dynamic flow of information between brain regions.