While a number of sophisticated computational and theoretical models exist for human behavior in the cognitive science literature, the relationship between these models and the underlying neural computation has rarely been explored. The aim of this thesis is to propose
novel applications of neuroimaging analysis methods combined with explicit modeling to bridge the gap between computational models and cognitive neuroscience, specifically in studies of higher cognition. In Chapter 1, I provide the necessary methodological background
to these projects, describing in detail current univariate and multivariate approaches to functional MRI (fMRI) analysis. I then describe three approaches, cross-classification, representational similarity analysis, and encoding analysis that allow claims about the underlying representations and computations to be made in neuroimaging studies. In Chapters 2, 3, and 4, I present experiments using these approaches, showing how they allow us to arbitrate between different theories of the representations and computations underlying higher cognition, building upon prior localization work. Finally, in Chapter 5 I propose a new computational framework for encoding analyses that allows for directly integrating computational models with neuroimaging analysis.