Natural fractures are an important storage space and major flow channels in tight sandstones, and have obvious multi-scale features. Thus the establishment of three-dimensional multi-scale fracture geological models has important practical significance for improving the development efficiency of this kind of reservoir. However, multi-scale fracture modeling methods lack systematic and detailed research. Especially the small scale fracture model generally has strong randomness and low reliability due to a single constraint condition. Therefore, taking the eighth member of the Yanchang Formation in wellblock 92 of the Honghe oilfield as an example, based on outcrop, core, logging and seismic data, this paper divides natural fractures into three types, namely large scale fractures, mesoscale fractures and small scale fractures. On this basis, a hierarchical modeling approach to multi-scale fractures is proposed: (1)The large scale fracture model is built by using the deterministic method of manual seismic interpretation; (2) the mesoscale fracture model is set up by using the deterministic method of ant tracking; (3)on the basis of establishing a 3D brittle index model and a seismic attribute fusion body, the multiple linear regression method is used to combine the rock brittleness index model, the fault distance property body and the seismic attribute fusion body to obtain the inter-well fracture development probability body, which is used as secondary data to establish the fracture density model.Next, under the constraint of fracture geometry and the fracture density model, the small fracture model is constructed by an object-based marked point processes simulation. Finally, a comprehensive discrete fracture network model is built by integrating the large scale, mesoscale and small scale fracture models and fracture properties models are established by upscaling. It is proved that fracture models established are in good agreement with the geological understanding and production data, which provide the geological foundation for the numerical reservoir simulation.