With the rapid development of automatic monitoring technology, it has become possible to carry out agricultural non-point source monitoring based on rainfall runoff events. The research on agricultural non-point source sampling strategies aims to improve the capture ability and efficiency of runoff events by setting reasonable sampling timing and frequency. This article systematically summarizes the research on agricultural non-point source output characteristics and automatic sampling strategies, and the results showed that: (1) small watersheds are the basic units of agricultural non-point source output, and pollutants usually exhibit pulse output characteristics. (2) The agricultural non-point source automatic sampling strategy based on rainfall runoff events includes basic elements such as base flow sampling frequency, runoff event initiation threshold, runoff event sampling interval, and sample mixing method. The base flow sampling frequency can be calculated through probability statistical formulas, and the initiation threshold is mostly based on real-time monitoring of water level or flow rate, and the sampling interval includes two types: time-interval (such as 15 min) and flow-interval(such as 5 mm). (3) The load estimation error of the sampling strategy is very sensitive to the startup threshold, and the startup threshold should not be set too high. The load estimation accuracy of flow interval sampling is usually higher than that of time interval sampling. The average deviation of the dispersed sample mode is lower than that of the mixed sample mode, and the average deviation of the mixed sample will expand with the increase of mixing quantity. (4) The sampling strategy should first consider the ability to capture runoff events (such as covering more than 90% of runoff) and the accuracy of pollution load estimation (with an error of less than 20% from the true flux). Monitoring objectives, watershed hydrological characteristics, and production and living factors also need to be comprehensively considered. At present, sampling strategies based on rainfall runoff events are mostly based on ideal runoff curves, which may have certain limitations in the application of complex hydrological and water quality change processes such as multi peak runoff and asynchronous water quality. In the future, sampling strategies can be continuously optimized and improved in the refinement of water quality change processes, deep coupling of watershed hydrological models, and real-time updating of environmental element information.