Vegetation cover is an important indicator of regional ecosystem change. Vegetation coverage is a synthetically quantitative index of conditions of vegetation community cover and an important characteristic variable of ecosystem models, water and carbon cycles models. Conventional vegetation coverage means the integrated results of different vegetation type, including tree, shrub and grass. When vegetation vertical heterogeneity is considered, decomposition of vegetation cover into tree and shrub/grass components using remotely sensed data is a new research field and will provide more ecological meaning parameters for quantifying the ecological environment and global climate change. Currently, a series of algorithms have been successfully used to retrieve tree, shrub and grass cover of horizontal scale from remotely sensed data, including vegetation indices, regression analysis, classification and regression tree, artificial neural networks, pixel unmixing analysis, physically-based model inversion, etc. These methods could meet the requirements of application accuracy. With the development of the new sensors, like LIDAR, multi-angle sensors as well as physically-based models, such as geometric optical and radiative transfer models, especially, two-layer canopy reflectance model, the retrieval of tree and shrub/grass cover of vertical scale in different temporal and spatial scale shows promising expectations. The paper reviews in detail the latest achievements and frontiers of the horizontal and vertical scale retrieval of tree, shrub, and grass cover from remotely sensed data, compares and analyzes main methods and models. In the end, it discusses the existing problems of various methods and gives an outlook of future research directions.