Groundwater monitoring is an indispensable approach to earthquake prediction in China. However, there are many different time-scales of nonlinear interferences in the observation data, by restricting the condition of most monitoring well. So, it is necessary to identify and eliminate these interferences by using advanced signal processing method. In this study, we take Wanshan well in Hubei province as an example, present a method of micro-dynamic response signal reconstruction based on EMD and frequency spectrum analysis. The result shows that water level date can be decomposed in 12 IMF components with definite physical significance, including earth tide(IMF1~5), rainfall(IMF6~8), climate(RES), and tectonic activity(IMF9~12). Finally, we reconstruct the water micro-dynamic signal by using IMF9~12, and discuss the relationship between seismic activity and the micro-dynamic signal from Jan.1 2014 to Nov.1 2015. The result indicates that the micro-dynamic of Wanshan well responding to the seismic activity around BeiheXiangfan-Guangji faults synchronously and obviously. The method can be prospective used in precursory anomaly discriminate and earthquake trend predict.