Abstract: The problem of organic pollution in river and lake water bodies in China is relatively significant. The assessment of organic pollution in rivers and lakes mainly employs chemical oxygen demand (COD) as the core indicator. COD source apportionment can identify the sources of organic pollution and determine the contributions of different pollution sources. By using appropriate methods for COD source apportionment, the causes of organic pollution can be clarified, which is of great significance for guiding the treatment of water pollution. This paper briefly analyzed the COD and its spatial distribution characteristics of river and lake water bodies in China. By using bibliometric methods, the temporal evolution of research hotspots in COD source apportionment methods was sorted out, and the application characteristics, advantages, limitations, and required data types of various source apportionment methods were summarized. The results show that the COD of river and lake water bodies in China presents obvious spatial differentiation characteristics. The rivers and lakes with relatively high COD are mainly concentrated in the lakes of the Mengxin Lake area and the Yunnan-Guizhou Plateau Lake area, as well as the river basins of the Huai River, Songhua River, Liao River, and Hai River. Bibliometric analysis reveals that the source apportionment methods for COD in rivers and lakes are transforming from traditional models to machine learning. The current mainstream source apportionment methods can be classified into four categories: emission factor, receptor model, mechanism model, and machine learning methods. The emission factor method is the simplest to calculate, but the spatio-temporal resolution of the data source is limited, and the COD generation and discharge coefficients are insufficient. The receptor model method can quickly perform source apportionment, but the identification of pollution sources is subjective. The mechanism model method is relatively accurate in simulating the dynamic migration and transformation of complex pollution sources, but the modeling is complex, and there are many types of input data. Machine learning has the advantages of fast computing and high precision, and is suitable for handling multi-dimensional nonlinear relationships, but it cannot simulate the migration and transformation process of pollution sources. In view of the limitations of the current mainstream methods in the practice of COD source apportionment, it is proposed that the combined use of multiple models should be developed in the future. The integration of machine learning and big data, as well as the integration of gene mapping and new technologies, should be carried out from aspects such as model collaborative optimization and technological innovation to promote the integrated development of multiple technologies, so as to improve the accuracy and efficiency of COD source apportionment.