Abstract: Chemical Oxygen Demand (COD), as a core indicator for assessing the degree of water pollution, is subject to organic pollution in many water bodies due to human interference and climatic influence. The source of COD in large water bodies remains unclear. The pollution situation of COD in surface water, influencing factors and current research hotspots are briefly introduced. The application characteristics, advantages, limitations and future development directions of the calculation model, receptor model, mechanism model and machine learning in the analysis of surface water COD pollution sources were summarized and compared with emphasis. The main research findings are as follows: (1) COD pollution in China shows significant spatial specificity. The Inner Mongolia-Xinjiang Lake area is the most polluted area, and the Yunnan-Guizhou Plateau Lake area is the secondary polluted area. The basins of the Songhua River, the Huai River and the Hai River are heavily polluted areas. (2) At the methodological level of source tracing, the receptor model is applicable to scenarios where the data is limited but the source needs to be resolved quickly. However, when calculating the contribution of pollution sources, the assumption of a linear relationship must be followed. The mechanism model, also known as the "white box model", is suitable for lakes with clear hydrological and water quality processes. It requires a large amount of basic data and is relatively accurate in simulating the dynamic migration and transformation of complex pollution sources, but has a high computational cost. Machine learning demonstrates excellent performance in environmental simulation and prediction, with fast computing speed and high simulation accuracy, and stands out in handling multi-dimensional nonlinear relationships. At present, pollution source apportionment technology has become a core tool for the diagnosis of water quality exceeding standards. One of the obvious drawbacks is that it is difficult to directly verify the error between the apportionment results and the actual source contribution, because it is impossible to obtain the real source discharge information on a large scale. The existing verification method is to use the mechanism model to simulate the data of the source contribution and then solve it with the receptor model. The receptor model and machine learning generally have the common problem of insufficient mechanism interpretation. A variety of interpretable technologies are adopted to break through the mechanism obstacles of the "black box model", and multi-model coupling technologies are developed to achieve a dual improvement in traceability accuracy and mechanism analysis. This includes not only the combination of different principle models but also the combination of artificial intelligence technology and traditional source analysis technology. This will become an important development trend for the precise traceability of COD in surface water