Abstract: Purple soil is inherently characterized by a loose structure and high erodibility, and it is prone to developing desiccation cracks under the alternating action of rainfall and evaporation. These cracks significantly damage the integrity of the soil structure and further degrade its water-holding capacity. To systematically clarify the influence mechanism of initial cracks on the water-holding characteristics of purple soil, this study systematically prepared undisturbed ring knife samples with different initial crack parameters, including crack ratio, crack length, average crack width, and crack number. The soil-water characteristic curve (SWCC) of each sample was determined using the axis translation method, and the Fredlund-Xing model was adopted to conduct fitting analysis on the measured SWCC data. Meanwhile, grey correlation analysis was employed to quantitatively identify the influence weights of various crack parameters on the characteristic parameters of SWCC and the fitting parameters of the Fredlund-Xing model. Additionally, low-field nuclear magnetic resonance (NMR) scanning tests, including nuclear magnetic resonance T₂ spectra analysis and pore throat distribution analysis, were carried out to intuitively verify the correlation between soil pore distribution and crack development. The results indicated that the influence of initial cracks on the water-holding characteristics of purple soil was mainly concentrated in the low suction section of SWCC, which was manifested as rapid dehydration accompanied by a distinct "plateau period". In contrast, the initial cracks had minimal impact on the transition section and residual section of SWCC, where the water-holding characteristics remained relatively stable. Further analysis showed that the air-entry value of the soil was inversely proportional to the crack ratio and dehumidification rate. The Fredlund-Xing model exhibited excellent fitting performance for the SWCC of purple soil with initial cracks, with the determination coefficient generally higher than 0.85. Among the model parameters, α, which is closely related to the air-entry value, was inversely proportional to the crack ratio and dehumidification rate, while parameters n and c showed no significant linear correlation with any of the crack parameters. The grey correlation analysis results further revealed that crack ratio was the primary dominant factor affecting the water-holding capacity of purple soil (correlation degree > 0.67), followed by the average crack width, whereas the crack number had the weakest correlation with the water-holding capacity. From the perspective of mechanism, initial cracks destroy the internal structure of the soil, thus forming a dual-pore system composed of interconnected crack space and intact non-cracked soil matrix. In this dual-pore system, the proportion of large pores dominated by macro-cracks increases significantly, while the proportion of medium and small pores controlled by the soil matrix decreases relatively. The preferential water flow effect of the crack system in the low suction stage is the core cause of the significant reduction in the soil air-entry value and the occurrence of the "plateau period" in SWCC. The NMR scanning tests further confirmed the close correlation between pore distribution and crack development: the nuclear magnetic resonance T₂ spectra and pore throat distribution results intuitively verified the negative effect of cracks on soil pore distribution. By integrating macro water-holding characteristics, micro pore structure characteristics, and model fitting parameters, this study established a multi-scale verification system for SWCC under crack disturbance, and clearly clarified the dominant role of the crack network as a preferential water flow path in regulating soil hydraulic behavior. This research systematically revealed the influence law and intrinsic mechanism of initial cracks on the water-holding characteristics of purple soil, and can provide a solid theoretical basis and key parameter support for the rational design of soil and water conservation projects, scientific optimization of soil improvement schemes, and accurate setting of landslide disaster early warning thresholds in purple soil distribution areas.