Magnetic resonance sounding (MRS) for groundwater investigation has received much attention of geophysicists due to its higher resolution, more efficiency, more information and unique interpretation for hydrogeological investigations. It has shown continuous development towards a frequently used geophysical technique over the last decades. There has been significant improvements in forward modeling, inversion and interpretation during the last years. These improvements extended the application range of the method concerning solutions to hydrogeological tasks. But the applicability of MRS is highly limited due to bad signal-to-noise (S/N) ratio. In many cases, the expected MRS signals are only few tens of nanovolts and therefore often contaminated by significant high-level electromagnetic noise. MRS group of Jilin University has studied the noise and its characteristics, and found that Power-line harmonic noise and spike noise are the most disturbing noises. The purpose of this paper is to give an overview of noise cancelling in MRS signal processing. We report the recent researches, including noise cancelling in real time, independent component analysis, spikes removal of MRS data based on energy calculation, as well as synchronous compression wavelet transform and self-optimizing nonlinear threshold compensation. By introducing several case studies, we prospect the developing trends of MRS noise cancelling.