Measuring subtle neuromagnetic signals requires eliminating background noises. Especially, a SQUID magnetometer is very sensitive to the magnetic noise even from a distant source. As typical software methods, we use the synthetic gradiometer of the adaptive filtering to reduce the noises. In this article, we present noise reduction effects in our 37-channel SQUID magnetometer system by applying each method including the frequency-domain adaptive filtering and discuss a selective application of the methods to the detection of clinical magnetoencephalogram signals.