Abstract: Many water diversion irrigation projects were launched in the arid areas of northwest China in recent years. Intense human activities have changed the water cycle of “diversion-irrigation-return” in the irrigation areas. In this study, the “Unit Return-Flow-Graph” was defined as the curve of irrigation return flow weight formed by deep percolation with uniform temporal and spatial distribution in a given watershed within a unit period. The Unit Return-Flow-Graph was then combined with the “Bucket model” (water balance model for crop root zone), thereby establishing the calculation model for the irrigation return flow. Deep percolation was also evaluated under the water balance, and then the Unit Return-Flow-Graph was combined to calculate the irrigation return flow. The study area was set as the Jingdian Irrigation District (part of the Yellow River Basin) in Gansu Province, China. The results showed that the calculated value of irrigation return flow in the study area fitted well with the monitored. The determination coefficient R2 in the model calibration and validation period were 0.82 and 0.71, respectively, indicating the reliable performance of the models. The validated model was used to calculate the deep percolation in the study area from 2000 to 2019 under the irrigation return flow from 2002 to 2019. The calculation results showed that the main influencing factor of yearly deep percolation was the sum amount of net irrigation and effective rainfall. The correlation coefficient between the sum amount of net irrigation and effective rainfall was 0.718 (P<0.01), indicating a significantly positive correlation. The main influencing factor of monthly deep percolation was the crop growth period in the irrigation area, where 69.6% of deep percolation occurred during winter irrigation (October to November). A large amount of irrigation water and rainfall were consumed by crops in the form of evapotranspiration during the crop non-growth period, with less deep percolation. The coefficient of yearly deep percolation was significantly positively correlated with the yearly deep percolation (r=0.944, P<0.01). The monthly coefficient of deep percolation was significantly dependent on the crop growth. It was less than 0.4 in the crop growing period, but greater than 0.8 in the non-growing period. The reason was that the crop consumed more water during the growing period, but less for deep percolation. Low water consumption but more deep percolation occurred in the crop non-growing period. The main influencing factor was the yearly deep percolation, where the correlation coefficient between the two was 0.716 (P<0.01), showing a significant positive correlation. The monthly irrigation return flow was correlated with the monthly deep percolation. The reason was that there was a significant lag time in the process of irrigation return flow. Since the curve was fitted to the Unit Return-Flow-Graph in the study area. The lagging peak of irrigation return flow was about 2 months, but the impact of deep percolation on the irrigation return flow reached about 24 months. The parameters of Unit Return-Flow-Graph presented clear physical meanings, relatively easy to determine the parameters using measured data. The Unit Return-Flow-Graph was effectively utilized to calculate the amount of irrigation return flow in water diversion irrigation areas, particularly on the water resources management in irrigation areas. In addition, the yearly and monthly deep percolation and irrigation return flow changed significantly, which affected the irrigation effect in Jingdian Irrigation District. The findings can provide a sound potential reference for water diversion in the irrigation districts of arid areas.