In the study of magnetoencephalography, it is important to obtain evoked fields with good signal-to-noise ratios (S/N) and with a small number of epochs in averaging. The noises are considered to be mainly spontaneous neuromagnetic fields. In the present study, we propose a method to improve the S/N. The basic principle of this method is the elimination of a principal component (PC) of multichannel-recorded neuromagnetic fields, utilizing the synchronized characteristics of spontaneous rhythmic activities dominating the fields. The proposed method is, therefore, called the principal component elimination method (PCEM). PCEM was applied to neuromagnetic fields measured by a 37-channel superconducting quantum interference device system, on which computer-generated evoked fields were superposed, in order to examine possible improvement in S/N. It was found that elimination of the first PC could improve the S/N of the evoked fields. The improvement in S/N with elimination of the first PC, compared to conventional simple averaging, increased with increases in the number of epochs and reached more than 50% after averaging over 128 epochs. PCEM also reduced the number of epochs needed in averaging to about half of that needed in conventional simple averaging.