Neural power spectra in the human brain consist of rhythmic oscillations, transient bursts and background 1/f like activity. Several cognitive processes have been linked to neural oscillations, which are thought to be a central mechanism of information flow across brain regions. In this research, we develop a new method based on source reconstructed magnetoencephalography (MEG) resting human brain data to isolate rhythmic episodes from the neural signal. To ensure the detection of genuine oscillatory activity across brain volumes, only local maximum peaks in the power spectra are selected. Peak power values are then compared against an empirical null distribution with no oscillatory components to calculate a power threshold. Only oscillations exceeding the power threshold (p<0.05) as well as a duration threshold (>3 cycles) are subsequently considered. Finally, we only take into consideration oscillatory activity in voxels showing spatial local maxima, thus removing the undesired source leakage effect resulting from the reconstruction algorithms. Here, we show the most representative brain regions of each canonical frequency band in a group of 128 participants from the Open MEG Archive (OMEGA). Results show an organized representation of rhythmic electromagnetic activity during the resting state. Critically, this approach allows for the first time the characterization of oscillatory episodes at the voxel level. Future research on how episodes of rhythmic activity of different neural generators relate to each other may provide critical information on brain communication.