This paper reviews a particular class of MEG source reconstruction techniques referred to as the spatial filter. The spatial filter can provide the spatio-temporal reconstruction of neural activities from MEG measurements. Two kinds of spatial filter techniques exist: one is a non-adaptive technique and the other is an adaptive technique. In the conventional non-adaptive technique, its spatial resolution has an intrinsic limitation imposed by a hardware configuration- An adaptive technique, however, can attain a spatial resolution that exceeds this hardware limitation. After reviewing these properties of the spatial filter techniques, this paper particularly focus on our proposed adaptive vector spatial filter technique, and shows that it can reconstruct spatio-temporal pattern of neural activities with a several-milli-meter spatial resolution and with a millisecond temporal resolution.