Abstract: Non-point source phosphorus (P) loss from farmland is one of the most serious causes of agricultural non-point source pollution. It is very necessary to identify the critical source areas and influence factors for the risk of P loss from farmland in a watershed, in order to prevent non-point source pollution. The objective of this study was to assess the risk of P loss from farmland in China from 2000 to 2020. P index model was also used. Among them, the soil available P content and fertilizer-P application rate were selected as the source factors. The soil-erosion modulus, annual runoff depth, and the normalized differential distance index between farmland and river network were used as the transport factors. Additionally, the GIS technology was then combined to identify the critical source areas of P loss from farmland. Random Forest (RF) was utilized to derive the critical influencing factors on the P loss from farmland in China. Structural Equation Modeling (SEM) was constructed to explore the relationship between the P index and influencing factors. The results show that: 1) The low, medium, high, and very high-risk areas of P loss from 2000 to 2020 accounted for 43.8%, 40.5%, 13.4%, and 2.4% of the total area of farmland, respectively. 2) The annual average percentage of the total area at high and very high risk of P loss from farmland in 2000, 2005, 2010, 2015, and 2020 was ranked in the descending order: the Huaihe River Basin, Yangtze River Basin, Pearl River Basin, Southeast River Basin, Songhua and Liaohe River Basin, Southwest River Basin, Yellow River Basin, Continental River Basin, and Haihe River Basin. 3) The RF results showed that the available P content and normalized differential distance index were the critical influencing factors of the P index, whose importance eigenvalues were 129.53 and 65.12, respectively. The available P content was the critical influencing factor of the P loss from the farmland. 4) SEM images showed that the P index was extremely significantly positively correlated with the source and transport factor indexes. The P index of the 14 selected index factors amounted to 0.62, in which the contribution rates of the source factor and the migration factor to the P index were 0.77 and 0.19, respectively (P<0.001). In conclusion, the findings can provide scientific references for the evaluation of non-point source pollution in farmland. It is of great significance for the decision-making on the prevention and control of agricultural surface pollution.