Soil is a complex with high heterogeneity. The early research on digital soil mapping mainly focused on the lateral variation of soil, with less consideration of the vertical variation and three-dimensional (3D) digital soil mapping. In recent years, the rapid developments of 3D geographic information technology and earth observation and detection technology have greatly promoted research on soil 3D data acquisition, 3D prediction, 3D data modeling, 3D model and visualization. In this paper, we reviewed the existing research on soil prediction and soil model construction in 3D space, to provide suggestions for the application and development of 3D digital soil mapping. We searched the Web of Science database by using 3D soil mapping, 3D GIS, 3D data model, 3D geological modeling, 3D visualization, soil spatial variability, spatial prediction, Kriging interpolation, soil-landscape analysis, depth function, machine learning, geostatistics, random simulation as keywords, and selected the key literatures for analysis based on correlation, citation rate and literature sources. We summarized the popular methodologies for soil spatial variability, 3D spatial soil prediction, soil 3D data model, and 3D model construction, and evaluated the advantages, disadvantages and application scenarios of each method. This review presents the common problems of 3D soil mapping, such as sparse soil profile data, low accuracy of 3D soil prediction, and insufficient information to create the data source for 3D soil modelling, and put forward some feasible research prospects.