An analysis of the characteristics of ultra-low-permeability reservoir rocks and the processes for permeability loss form the basis for establishing high-resolution well log interpretation models. Moreover, this type of analysis is a prerequisite for the effective development of ultra-low-permeability reservoirs. The Chang 7-1 sandstone members of the Triassic Yanchang Group in the Longdong region of the SW Ordos Basin are typical ultra-low-permeability reservoirs and are studied in detail in this paper. A comprehensive methodology of evaluating the ultra-low-permeability reservoir is developed based on geological information and well logging data. First, pore structure characteristics are analysed in depth using core observations, cast thin sections, scanning electron microscopy and mercury injection porosimetry. The mechanism behind the formation of ultra-low permeability in a reservoir rock, and its controlling factors, are also clearly determined. Second, from the above results and the pore structure characteristics in the study area, the flow-zone indicator (FZI) is selected as the key reservoir modelling parameter to establish permeability models that can reflect different pore structure types and improve precision compared to non-classification pore structure models. In addition, conductivity experiments with water-saturated cores confirm that the additional conduction phenomenon of clay minerals has little impact in terms of reducing reservoir resistivity and, consequently, the Archie formula is found to be reliable for ultra-low-permeability rocks with porosities greater than 5% in the Chang 7-1 reservoir. Three types of well log interpretation models, such as permeability, oil saturation and porosity, are established for the ultra-low-permeability reservoirs using FZI as a classifier in the research area. These models reduce the relative errors to less than 10% and performed well for the ultra-low-permeability Chang 7-1 sandstones. This study contributes to the evaluation of ultra-low-permeability reservoir rocks and also generates fit-for-purpose log evaluation models to guide completion interval selection.