For unconventional reservoir hydraulic fracturing design, a greater fracture length is a prime factor to optimize. However, the core observation results from the Hydraulic Fracturing Test Site (HFTS) show that the propped fractures are far less or shorter than expected, which suggests that the roughness and tortuosity of hydraulic fractures are crucial to sand transport. In this study, a transport model of sands is first built based on experimental measurements on the height and transport velocity of the sand bank in fractures with predetermined width and roughness. The fracture roughness is quantified by using the surface height integral. Then, three-dimensional simulations are conducted with this modified model to further investigate the impact of tortuous fractures on sand transport, from which a regression model is established to estimate the propped length of hydraulic fractures at a certain pumping condition. The experiment results show that the height of the sand bank in rough fractures is 20–50% higher than that in smooth fractures. The height of the sand bank decreases with the reduction in slurry velocity and increases with the increase in sand diameter. Sand sizes do little effect on the transport velocity of the sand bank, but the increase in slurry velocity and sand volume fraction can dramatically enhance the migration velocity of the sand bank. The appearance of tortuous fractures decreases the horizontal velocity of suspended particles and results in a higher sand bank compared with that in straight fractures. When the sand bank reaches equilibrium at the tortuous position, it is easy to produce vortices. So, there is a significant height of sand bank change at the tortuous position. Moreover, sand plugging can occur at the entrance of the fractures, making it difficult for the sand to transport deep into fractures. This study explains why the propped length of fractures in HFTS is short and provides a regression model that can be easily embedded in the fracturing simulation to quickly calculate dimensions of the propped fractures network to predict the length and height of propped fractures during fracturing.