Effective event picking is the basic and critical link in microseismic data processing, but the picking effect of traditional picking methods depends heavily on the selection of picking parameters, and it is easily affected by signal characteristics and signal-to-noise ratio, which is difficult to meet the requirements of real-time processing of massive monitoring data. Based on this, in order to effectively meet the current processing needs of large-area, wide-azimuth and high-density monitoring data, based on the structural characteristics of NARX neural network, a network model that meets the needs is constructed, and the preprocessed synthetic and measured microseismic signals are fed into its Series-Parallel feedback structure for training to fully learn the waveform characteristics of the signals. Then, the monitoring signal to be picked is simply processed and inputted into the NARX model with good performance to output the characteristic curve, and the effective event is picked by the threshold. The results of picking of synthetic and measured microseismic signals of the test database and the comparison with the STA/LTA method show that the NARX method can complete the picking of effective events of microseismic signals, and it shows strong noise resistance and adaptability, and it has certain advantages over STA/LTA method, so it has the potential to become one of the effective tools for picking the events of microseismic monitoring signals.