Standard statistical modeling for functional magnetic resonance imaging (FMRI) during task performance or in a resting state typically involves implementing a voxel-wise general linear model at the single-subject level to identify regions of the brain that are modulated by task performance or that are functionally connected with a seed region of interest during rest. Multi-subject analysis of resulting regression parameter estimate spatial maps are then conducted using a voxel-wise higher-level GLM. While these methods are powerful approaches for hypothesis testing, there are several limitations of such approaches, including discarding spatial information during modeling, requirement to specify an accurate model of brain activity, and/or to specify a seed-region of interest. In this talk, I will discuss recent work on multivariate and data-driven methods for statistical analysis, techniques that obviate some of these concerns. Approaches discussed will include group independent component analysis (ICA) with dual regression to study brain network functional connectivity and multivariate spatial regression to study brain network task activation. I will also discuss our new technique for denoising scanner and study effects from multi-site data that utilizes a data-driven approach for multi-modal MRI data fusion called linked ICA.