The U.S. Geological Survey Parkfield Dense Seismograph Array (UPSAR) successfully recorded strong motions during 2003 San Simeon earthquake (M 6.5) and 2004 Parkfield earthquake (M 6.0). Because the array covers a very small area (0.45km2), these data offer some interesting insights into spatial variations of seismic ground motions that suits for engineering scale. In this research, we study the spatial coherency function of seismic ground motion in the horizontal and vertical directions by digital signal processing. The results show that when the circular frequency is smaller than , the degressive trend of the coherency function becomes significant with the separation distance elongation, while the data deviation of the coherency function becomes larger with the frequency rise, which shows no obvious rules. In addition, based on the strong-motion data, a suitable spatial coherency model of ground motion is selected through comparing existing model functions, and the appropriate recommendations for improvement is put forward. Finally, according to different frequency range, the fitting parameters of the spatial coherency function of ground motion are obtained through numerical simulation. The spatial coherency function proposed in this paper is practical in simulation of ground motion field.