To separate the EEG measured on the scalp into the parts generated by deep and shallow dipole sources respectively is a subject of great significance both theoretically and clinically. Based on the analysis of the spatial frequency of EEG

a method is proposed in this paper to solve this problem in two steps.For the first step

the shallow dipoles are localized by the higher spatial frequency components and for this purpose

a frequency domain Marquardt algorithm is derived.Then

in the second step

the electrical activities contributed by the deep dipoles are revealed after substraction of the potentials generated by shallow dipoles from the measured EEG. Computer simulations show that the shallow sources can be localized accurately by inversion in the frequency domain with higher frequency components and the potentials generated by deep and shallow dipoles can be separated effectively.