Taking a residential area in Guangzhou as the research object, rainwater runoff from roof, main road and green area was monitored and sampled during rainfall to analyze the characteristics of pollutant concentration changes over time and the correlation between different pollutants in the study area. Based on the mean event mean concentration (EMC) of pollutants as the evaluation standard, a three-layer BP neural network model is constructed, which can predict the pollution load in stormwater runoff of residential areas in Guangzhou after training, calculation and testing. The results show that the pollution degree of stormwater runoff in the main road of the residential area is relatively serious. With the increase of rainfall, the peak time of the mass concentration of various pollutants is advanced and the value is increased, and the EMC value of pollutants in stormwater runoff generally presents a trend of first increasing and then decreasing. The chemical oxygen demand (COD) and total phosphorus (TP), total nitrogen (TN) and suspended solids (SS) in different types of subsurface runoff showed strong correlation. After training with the cross-validation method, the average relative error of the BP neural network model for predicting the EMC value of different pollutants in the test set is < 10%. Therefore, it can be considered that the model has certain reference value for predicting the stormwater runoff pollution load in residential areas of Guangzhou.