With the use of CT scanning to construct a 3D digital image model of Gaomiaozi bentonite, the reconstruction accuracy of the pore structure was improved through optimized image processing methods. By combining statistical results from high-resolution 2D SEM images, the representative elementary volumes of the 3D image model were extracted using fractal dimension and porosity as constraints. A pore network model was constructed via the maximum sphere method, and the pore structure characteristics and their effects on seepage properties were quantitatively analyzed in terms of geometric morphology, topological structure, and spatial homogeneity. Results show that: the pore radii of the studied Gaomiaozi bentonite samples mainly concentrate in the range of 45–105 μm, showing an approximately normal distribution; throat radii smaller than 45 μm account for 90%, and the regional shape factor is concentrated in the range of 2–15. Pores with a coordination number of 1–4 account for approximately 60%; pores with Euler 3D values of 1 and 0 are more concentrated, indicating relatively simple topological structures. Units with a pore-throat ratio of 1–4 account for about 70%; pore flatness and elongation show concentrated distributions, indicating good spatial homogeneity. Permeability spans multiple orders of magnitude (1.0×10^?16–1.0×10^?21 m²), with approximately 73% of pore-throat units belonging to the dominant flow region (1.0×10^?16–1.0×10^?18 m²). About 15% of pore channels show no seepage contribution, whose critical pore diameters are consistent with the throat radius distribution and negatively correlated with the pore-throat ratio. Coordination number distributions for different critical pore diameters are also similar.