Recognizing others enables us to acquire knowledge about them, and to retrieve this knowledge in future encounters. Understanding the neural basis of the recognition of person identity requires 1) individuating which brain regions are engaged during recognition, 2) characterizing the representations they encode, and 3) modeling how these representations are computed from the inputs they have available (representations encoded in afferent regions). I investigate these questions with a combination of 1) univariate and 2) multivariate fMRI analyses, and 3) introducing a new approach we developed that unifies multivariate pattern analysis and connectivity methods to model nonlinear mappings between multivariate representations encoded in different brain regions. I then discuss the applicability of this approach to other domains, using the case of language as an example.