Coseismic signals of groundwater levels are generally obtained by subtracting the responses of atmospheric pressure, Earth tides and precipitation from the observed data. However, if the observations are conducted without nearby barometers, pluviometers or Earth tide data for correction, coseismic signals are often difficult to extract. In this case, the Hilbert-Huang transform (HHT) is used to obtain the instantaneous frequencies and amplitudes for every point of the decomposed intrinsic mode functions (IMFs) from groundwater level data for differentiating the related frequency-dependent responses without a further auxiliary input. The extracted coseismic signals show intense amplitude pulses that are clearly seen in the third IMF. In addition, two types of coseismic signals can be readily distinguished in the results from the HHT transform. One is an instantaneous short-time signal induced by the passing of seismic waves. Another coseismic signal is a sustained signal induced by the near-field earthquake occurring near the Hualien station, Taiwan and shows a positive correlation between the earthquake distance and magnitude. Our results generally show that using the HHT transform improves our understanding of automatic detection of the coseismic signals from the groundwater level.