Abstract: Accurate identification and quantification of in-river non-point source pollution flux represents a critical scientific link in the precise prevention and control of agricultural non-point source pollution. Current research and practice predominantly focus on "source-to-sink" process simulation at the watershed scale, while lacking direct quantitative analysis methods based on cross-section monitoring data for "sink-to-source" tracing. This study employed bibliometric methods to systematically review 4950 relevant articles from the Web of Science Core Collection spanning four decades (1980-2025), constructing a comprehensive knowledge map of calculation methods for surface direct runoff non-point source pollutant flux at river cross-sections. Building on this foundation, using the Ciba cross-section of the Linjiang River Basin in Yongchuan, Chongqing as an empirical case, we integrated high-frequency monitoring data with methodological findings from the literature to comparatively evaluate multiple pollutant flux separation techniques. The core of this research lies in the high-precision decomposition of the measured total pollutant flux at river cross-sections into three components: baseflow (groundwater background value), point sources (centralized discharges), and surface runoff non-point sources (primarily agricultural non-point sources), thereby achieving "observation-based source tracing" direct quantification of non-point source pollution contributions. The findings of this study provide a highly operational methodological system and typical case support for the precise identification, responsibility attribution, and load accounting of non-point source pollution in water environment management, holding significant scientific value for promoting the paradigm shift of water quality management in China from concentration control to flux control.