In standard fMRI analysis all voxels are tested in a massive univariate approach, that is, each voxel is tested independently. This requires stringent corrections for multiple comparisons to control the number of false positive tests (i.e., marking voxels as active while they are actually not). As a result, fMRI analyses may suffer from low power to detect activation, especially in studies with high levels of noise in the data, for example developmental or single-subject studies. Activated region fitting (ARF) yields a solution by modeling fMRI data by multiple Gaussian shaped regions. ARF only requires a small number of parameters and therefore has increased power to detect activation. If required, the estimated regions can be directly used as regions of interest in a functional connectivity analysis. ARF is implemented in the R package arf3DS4. In this paper ARF and its implementation are described and illustrated with an example.