We combine light detection and ranging (LiDAR) digital terrain model (DTM) data and an improved mesh implementation to investigate the effects of high-resolution surface topography on seismic ground motion based upon the spectral-element method. In general, topography increases the amplitude of shaking at mountain tops and ridges, whereas valleys usually have reduced ground motion, as has been observed in both records from past earthquakes and numerical simulations. However, the effects of realistic topography on ground motion have not often been clearly characterized in numerical simulations, especially the seismic response of the true ground surface. Here, we use LiDAR DTM data, which provide two-meter resolution at the free surface, and a spectral-element method to simulate three-dimensional (3D) seismic-wave propagation in the Yangminshan region in Taiwan, incorporating the effects of realistic topography. A smoothed topographic map is employed beneath the model surface in order to decrease mesh distortions due to steep ground surfaces. Numerical simulations show that seismic shaking in mountainous areas is strongly affected by topography and source frequency content. The amplification of ground motion mainly occurs at the tops of hills and ridges whilst the valleys and flat-topped hills experience lower levels of ground shaking. Interaction between small-scale topographic features and high-frequency surface waves can produce unusually strong shaking. We demonstrate that topographic variations can change peak ground acceleration (PGA) values by ±50% in mountainous areas, and the relative change in PGA between a valley and a ridge can be as high as a factor of 2 compared to a flat surface response. This suggests that high-resolution, realistic topographic features should be taken into account in seismic hazard analysis, especially for densely populated mountainous areas.