Terrain analysis is based on a set of methods and techniques developed since the 19 Century addressing a broad range of application domains. Classification is a key scientific approach to extracting information and ultimately knowledge from data collections. Terrain classification prepares information from digital terrain models for various application scenarios. Spatial segmentation is a relatively novel approach based on classification and regionalization. This chapter is intended as a review of current practice with an emphasis on providing the basic conceptual foundations for segmentation in terrain analysis. It starts by linking segmentation approaches to basic geographical concepts such as spatial categories vs. regions, the process of regionalization, concepts of scale, interfacing with human perception, and physical spatial processes. Segmentation attempts to combine powerful characteristics from the continuous field view with spatial object/topographic type/landscape unit views. With an ever increasing number of sensors and terrain data acquisition technologies, terrain analysis should no longer be driven by traditional data-centric approaches, but rather by semantics-centred concepts from the respective application domains. Only by identifying and delineating elementary spatial units independently from data sources, can physical models become truly interoperable and transferable across different types of terrain data sets.