We present a comparison of three methods for the solution of the mag- netoencephalography inverse problem. The methods are: a linearly constrained minimum variance beamformer, an algorithm implement- ing multiple signal classification with recursively applied projection and a particle filter for Bayesian tracking. Synthetic data with neurophys- iological significance are analyzed by the three methods to recover po- sition, orientation and amplitude time course of the active sources. Finally, a real data set evoked by a simple auditory stimulus is consid- ered.