An automatic detection algorithm of P-wave arrival time for a microearthquake is described. A function based on normalized fourth-order cumulant, kurtosis, is used to detect both provisional P-wave arrival time and the end of microseismic signal. The contrast function has the maximum value near the P-wave arrival time. After detection of the provisional P-wave arrival time, maximum likelihood analysis is applied to evaluate residuals between microseismic signal and its AR model, which is made based on the provisional P-wave arrival time. 100 of microseismic events are used to examine the performance of this automatic detection. Errors of the automatic detection are within 3 ms in the 97 events. 3 ms is as long as a quarter of a period of P-wave because dominant frequency of P-wave is 80Hz.