Low-permeability sandstones have strong heterogeneity and complex pore structure, making it difficult to construct high-precision permeability evaluation models with conventional methods. To improve evaluation accuracy, 321 core samples from the Wenchang Formation in the Huizhou area, eastern South China Sea Basin, were tested by nanoscale computed tomography (CT) and nuclear magnetic resonance (NMR). Micron-scale porosity was identified as the key factor controlling permeability. A novel permeability prediction model was established based on micron-scale porosity. Using two T₂ cutoff values, the NMR T₂ spectrum was divided into nanoscale, micron-scale and submillimeter-scale pore components, to realize quantitative calculation of micron-scale porosity. The average relative error between calculated micron-scale porosity and CT measurement is 13.24%. The permeability error of the new model is only 24.07%, much lower than 50.27% of the traditional porosity-permeability model, which proves the new model has higher accuracy and applicability.