Neural source localization using electroencephalographic data is usually performed using either dipolar models or minimum norm based techniques. While the former demands a priori information about the number of active sources and is particularly suitable for generators, which occupy small pieces of cortical tissue, the major drawbacks of the second approach are its dependence on the uncorrelated noise, and its tendency to localize the sources at the surface. In this paper, a simple mathematical procedure, based on the behavior of the dispersion of the minimum norm solutions, is introduced, in order to estimate the depth of the sources. The correct position of the active generators is obtained using successively deeper surfaces instead of the application of a regularization matrix, as is commonly described in the bibliography. The evaluation of this technique is performed using single and double dipolar simulated generators and two different models for the head: spherical and realistic. The results yield a mean accuracy of about 10 mm for the most disadvantageous situations studied and thus, this method seems to be very promising to handle the depth of the neural generators.