Abstract: Lagged coherence is an important indicator to characterize the spatial variations of earthquake ground motion. In order to study the spatial variation and their differences of lagged coherence between near-field and far-field ground motion, the analyzed is the horizontal and vertical coherence functions of two dense arrays based on the near-field record of the 2004 Parkfield M_w6.0 earthquake recorded by the UPSAR array in the United States and the far-field record of the 2008 Wenchuan M_s8.0 earthquake recorded by the Zigong array in China. The horizontal and vertical coherence functions of near-field and far-field ground motion were analyzed separately as a function of distance and frequency, and the differences in lagged coherence were compared. The lagged coherence values based on actual data are fitted and analyzed with three commonly used coherence models to reveal the advantages and limitations of different coherence models. The results show that: there is a significant difference in the decay rate of coherence between near-field and far-field events, and the coherence of far-field recordings is remarkably greater than that of near-field recordings at low frequencies and near distances; coherence decays at different rates with distance and frequency, and the coherence function model requires different parameter descriptions for the frequency and distance terms. The existing models differ in their fits to near-field and far-field ground motion, with better fits to far-field records than to near-field records. The three models fit the same event differently, with the Harichandran model working best, followed by the Feng model, and Luco model being less applicable.