One of the problems in analyzing magnetoencepharograpy (MEG) is that brain signals are contaminated with high-level noise and artifacts. Although independent component analysis (ICA) is a useful method to separate brain signals from other components, not all signals are statistically independent. Additionally, each component should be judged as a brain signals or the others objectively. In this paper, we propose two ICA approaches that utilize spatial characteristics of brain activities to separate signals more precisely and meaningfully. Numerical experiments showed that it is helpful for ICA to use spatial arrangement, and a experiment using auditory evoked field (AEF) data brought out the features of proposal techniques.