A major goal for fMRI research is to figure out how people's brains differ in ways relevant to some phenotype of interest (e.g., disorder status). But there are serious challenges because the brain has tens of thousands of voxels (or connections) and each is measured with substantial noise. Our aim in this meeting is to introduce a PCA-based method we have developed, called Brain Basis Set, for predicting phenotypes of interest. We will also discuss some theoretical reasons why PCA-type methods have advantages over alternatives for individual differences fMRI research.