Using continuous 1-Hz sampling time-series recorded by a SG(superconducting gravimeter) at Hsinchu station, Taiwan of China, we investigate the anomalous gravity signals prior to 71 large earthquakes with moment magnitude larger than 7.0(Mw7.0) occurred between 1 Jan 2008 and 31 Dec 2011. We firstly evaluate the noise level of the SG records at Hsinchu(HS) station in microseismic bands from 0.05 Hz to0.1 Hz by computing the PSD(power spectral density) of seismically quiet days selected based on the RMS of records. Based on the analysis of the noise level and the spectral features of the seismically quiet SG records at HS station, we detect AGSs(anomalous gravity signals) prior to large earthquakes. We apply HHT(Hilbert-Huang transformation) to establish the TFEP(time-frequency-energy paradigms) and MS(marginal spectra) of the SG data before the large earthquakes, and the characteristics of TFEP and MS of the SGs data during the typhoon event are also analyzed. By comparing the spectral characteristics of the SGs data during seismically quiet period, three types of AGSs are found; and the occurrence rate of AGSs before 71 earthquakes is given in terms of the cases with different epicenter distance and different focal depth. The statistical results show that 56.3% of all the examined large earthquakes were preceded by AGSs; and if we constrain the epicenter distance to be smaller than 3500 km and focal depth less than 300 km, 75.3% of the examined large earthquakes can be associated with the AGSs. Especially, we note that for all the large earthquakes occurred in the Eurasian plate in recent four years, the precursory AGSs can always be found in the SG data recorded at HS station. Our investigations suggest that the AGSs prior to large earthquakes may be related to focal depth, epicentre distance and location. Using continuous 1-Hz sampling time-series recorded by a SG (superconducting gravimeter) at Hsinchu station, Taiwan of China, we investigate the anomalous gravity signals prior to 71 large earthquakes with moment magnitude larger than 7.0 (Mw 7.0) Jan 2008 and 31 Dec 2011. We first evaluate the noise level of the SG records at Hsinchu (HS) station in microseismic bands from 0.05 Hz to 0.1 Hz by computing the PSD (power spectral density) of seismically quiet days selected based on the RMS of records. Based on the analysis of the noise level and the spectral features of the seismically quiet SG records at HS station, we detect AGSs (anomalous gravity signals) prior to large earthquakes. We apply HHT (Hilbert-Huang transformation) to establish the TFEP (time-frequency-energy paradigms) and MS (marginal spectra) of the SG data before the large earthquakes, and the characteristics of TFEP and MS of the SGs data during the typhoon events are also analyzed. By comparing the spectral characteristics of the SGs data during seismically quiet period, three types of AGSs are found; and the occurrence rate of AGSs before 71 earthquakes is given in terms of the cases with different epicenter distance and different focal depth. The statistical results show that 56.3% of all the examined large earthquakes were preceded by AGSs; and if we constrain the epicenter distance to be smaller than 3500 km and focal depth less than 300 km, 75.3% of the examined large earthquakes can be associated with the AGSs. Especially, we note that for all the large earthquakes occurred in the Eurasian plate in recent four years, the precursory AGSs can always be found in the SG data recorded at HS station. Our investigations suggest that the AGSs prior to large earthquakes may be related to focal depth, epicentre distance and location.