We recently proposed a technique able to analyze the trend of the seismic activity by combining neural network model and seismic factors. In this paper, a variation of seismicity was introduced to reflect the corresponding vary of the frequency of earthquake. This variation was used as the precursor of future moderate earthquake. The time intervals of earthquake will be obtained through training the neural network. Then we judge the occurrence of strong earthquakes according to the time intervals. Finally, some statistical researches were made by using the method for the earthquake catalog in Sichuan and Taiwan in China. Through this above, we verify the validity of this method and state the general steps of this method.