Most statistical methods for assessing activated voxels in fMRI experiments are based on correlation or regression analysis. In this context the main assumptions are that the baseline can be described by a few known basis-functions or variables and that the effect of the stimulus, i.e. the activation, stays constant over time. As these assumptions are in many cases neither necessary nor correct, a new dynamic approach that does not depend on those suppositions will be presented. This allows for simultaneous nonparametric estimation of the baseline as well as the time-varying effect of stimulation. This method of estimating the stimulus related areas of the brain furthermore provides the possibility of an analysis of the temporal and spatial development of the activation within an fMRI-experiment.