Abstract: The broadband seismometer or gravimeter installed on the surface of the earth record not only the seismic wave signal excited by the big earthquake, but also the continuous weak high-frequency vibration signal, which is the earth background noise. Therefore, the research on the background noise of the gravimeter is helpful for the site selection of the station, the adjustment of the instrument, the judgment of whether the instrument has the ability to detect the normal mode of earthquake in the near future, and the detection of weak geodynamic signals. In order to evaluate the background noise level of PET/gPhone gravimeters of the National Gravity Network of China, we collected and processed 4-yr continuous gravity data recorded by the 52 PET/gPhone gravimeters from 2014 to 2017. The power spectral density during the quiet period of the gravimeter will be used as an estimate of the background noise level. Firstly, we compared and analyzed the existing background noise level calculation methods, and proposed an improved data selection strategy (called DaySNM method) for SNM (seismic noise level). Then we calculated background noise level at seismic frequency band (200～600 s) based on the Banka method(Banka and Crossley, 1999), presenting a new low-noise level model (CN-NLNM) by fitting the lower envelope of 52 power spectral density curves. Finally, the spatial distribution map of background noise in mainland China was obtained by Continuous Curvature Spline Interpolation with the SNM data. The results show that: (1) DaySNM method can find a lower SNM, compared with the Banka and Bandpass method. (2) the variation range of SNM in mainland China is 2.013 (Yutian station)～4.654 (Yinchuan station). (3) SNM of gravimeters in coastal areas generally larger than inland, indicating that ocean noise may be the main reason for the greater noise of coastal stations; (4) SNM in North China, eastern and northern Qinghai-Tibet Plateau is huge, which may be related to the frequent tectonic activities in these areas.