Abstract: The microseismic first arrival time picking is an important prerequisite for the high-precision positioning of the microseismic source. The traditional manual picking method is inefficient. The automatic picking method is difficult to pick the arrival time of the first wave accurately under the condition of low signal-to-noise ratio. In order to solve the above problems, an automatic picking method of microseismic first arrival time based on improved support vector machine (SVM) is proposed. Firstly, the method carries out normalization processing, linear correction and proper clipping on original microseismic data. The method marks different categories of the data by taking the amplitude, the energy and the energy ratio of adjacent moments of the microseismic data as features. Secondly, the method adopts a particle swarm optimization (PSO) algorithm and a grid search method to optimize the penalty parameters and the kernel function parameters of the SVM. The method carries out large-range fast positioning on the parameters by using the PSO algorithm to obtain a preliminary optimal solution. Then the method re-constructs a parameter search interval by taking the solution as an initial position, sets a small-step grid search method to carry out fine searching on the parameters to obtain the optimal parameters. The method substitutes the optimal parameters into the SVM model to train, and obtains the improved SVM model. Finally, the microseismic data are classified and identified according to the improved SVM model. The time corresponding to the first sampling point of the microseismic wave is defined as the arrival time of the first wave. The microseismic monitoring data from a mine shaft is used for the experiment. The results show that the accuracy of the method for picking the microseismic first arrival time is 96.5%, and the average picking error is 3.8 ms. Under the condition of low signal-to-noise ratio, the microseismic first arrival time can still be picked accurately. The picking precision is higher than the short term average/long term average (STA/LTA) method commonly used in automatic picking methods.