The development of tight oil reservoirs with volume-fractured horizontal wells under natural depletion faces the problems of rapid production decline and low primary oil recovery. Because the remaining oil potential in each fractured section of horizontal wells is unclear，the production performance of the horizontal wells after refracturing shows a significant difference. Determining the remaining oil potential of each fractured section in horizontal wells is of great significance to the design of the refracturing scheme and the improvement of the horizontal well production performance. Therefore，Block FX in Ordos Basin was taken as an example，and 12 factors were selected as input parameters of the back propagation（BP）neural network by analyzing the influencing factors of hydraulic fracturing-stimulated reservoir volume. As a result，the BP neural network prediction model for fracture length and fracture zone width was established. The prediction errors of seven test samples were 4.91% and 2.42%，respectively，which were relatively small，indicating that the prediction results of this model were accurate and reliable. On this basis，the remaining oil in fractured sections was quantitatively studied by using the volumetric method and the equivalent fractured well diameter model. The method was applied to the horizontal wells in the study area. According to the remaining oil potential evaluation results of each fractured section，the targeted refracturing measure was designed，and the field test was carried out，which achieved positive results and further verified the reliability of the method.