Abstract: Soil water retention curve (WRC) is closely linked to the pore distribution and soil structure in hydrological and agricultural practices. Loess caves have also led to substantial soil erosion and the destruction of arable land in the western region of Shanxi Province, China. The hydraulic properties of the soil can dominate the distribution of these loess caves. The WRC can provide both direct and indirect insights into these properties. In this study, CT technology and 3D image analysis were utilized to determine the hydraulic resources and their evolution in the loess at different depths within this region. The distribution of pore size was obtained in the loess samples through a series of image processing. The Young-Laplace equation was used to calculate the amount of water released from the samples under varying degrees of matric suction. The soil water retention was estimated after calculation. The van Genuchten (VG) model was then employed to fit the resulting WRC curves. Evaporation tests were conducted to determine an average residual saturation of 12.3 because the water content in the porous media failed below the residual water content of the substrate. After that, the estimation was verified to allow an actual drainage pore occupancy of 87.7%. The image analysis showed that the porosity was approximately 12% smaller than that measured physically, due primarily to the limitations of the sample size during CT scanning. Correspondingly, the correction factor was then applied to reduce the calculation errors. The pore sizes were adjusted to range mainly between 1.04 and 1.06. Pores larger than 20 μm in loess followed a log-normal distribution, accounting for over 70% of the total pore volume. These pores were the primary spaces for water retention and release. Furthermore, the pore structure varied significantly in the samples, as the depth increased from 2 to 8 m. The average size of pores decreased by 7.31% under compaction, indicating the broader distribution of pore size. There was a 45.55% increase in the air entry value with the smoother WRC curve. Both sphericity and aspect ratio were improved to reduce the permeability. Consequently, the thicker soil layers demonstrated better water retention. The porosity slightly decreased from the north to the south across the entire region, which was closely related to the decrease in the particle size of loess. The clay content increased by 163.82%, thus causing the average pore size to shrink by 18.13%, and the air entry value to rise by 267.24%. The increasing pore sphericity and decreasing aspect ratio further enhanced the capacity of water retention in soil. Most loess caves shared diameters ranging from 0.5 to 1.5 m and depths between 4 and 8 m. There were vertical variations in the WRC of the loess layers, the presence of slope-critical interfaces, and the widespread development of primary seepage channels, such as the vertical joints and unloading cracks. The material and spatial foundation were provided to form the loess cave erosion. These caves were formed to combine hydraulic erosion and gravitational deposition. The highly permeable loess layer also inhibited to form the caves in the northern, while the low permeability soil promoted the surface runoff in the southern. As such, the density of the cave increased and then decreased from the north to the south. In summary, μCT can provide a relatively simple and rapid approach to characterize the key features of soil hydrology and agricultural properties. The loess caves depended on various factors, such as hydraulic behavior and geographical location. While the WRC can offer important insights into the patterns of cave depth and density.