Terrain representation is one of the most common tasks in visualization of geoinformation. In order to realistically visualize the real landscape, several issues have to be addressed, in particular, how to combine terrain meshes and texture images. In this paper, Level of Detail (LoD) is used to treat both terrain meshes and texture images for producing realistic visualization with better performance but without too much information lost. Several layers of terrain grids and texture images with different levels of details created from original high resolution digital elevation model and satellite image using quad-tree and image pyramid algorithms, respectively, and to be used as base data. Then, new terrain meshes consisting of different LoD patches are generated dynamically according to the viewing position and viewing distance. Similarly, corresponding texture images with regions of different resolutions are also created instantaneously according to viewing condition changes.