The frequency of microseismic data is higher than that of conventional seismic data. The range of effective frequency is usually from 100 to 500 Hz, and low-frequency noise is a common disturbance in downhole monitoring. Conventional signal analysis techniques, such as band-pass filters, have their limitation in microseismic data processing when the useful signals and noise share the same frequency band. We have developed a novel method to suppress low-frequency noise in microseismic data based on mathematical morphology theory that aims at distinguishing useful signals and noise according to their tiny differences of waveform. By choosing suitable structure elements, we have extracted low-frequency noise from a original data set. We first developed the fundamental principle of mathematical morphology and the formulation of our approach. Then, we used a synthetic data example that was composed of a Ricker wavelet and low-frequency noise to test the feasibility and performance of the proposed approach. Our results from the synthetic example indicate that the proposed approach can effectively suppress large-scale low-frequency noise while slightly decreasing the small-scale signals. Finally, we have applied the proposed approach to field microseismic data and obtained very encouraging results.