In order to enhance the reliability of electroencephalogram (EEG) signal source analyses, utilizing EEG lead field matrices obtained by field analyses in custom-made real head models is effective technique. Custom-made models are usually constructed from voxel models derived from magnetic resonance (MR) images using a variety of image-processing techniques. We have improved one of the techniques that select threshold levels dividing signals and noises based on the EM (expectation-maximization) algorithm. This technique contributes to rapid and high-quality voxel model acquisition. We demonstrate the following operations: (a) voxel model construction, (b) lead field calculation, and (c) simulation of EEG electrode voltage measurement induced by an equivalent current dipole (ECD), which is set in the primary motor cortex by considering application of brain-machine interfaces. The proposed technique is compared with other techniques based on the voltage differences caused by the constructed model-shape differences.