The concentration characteristics of pollutants (represented by SS and TP) and the adaptability of prediction models of surface runoff in Pingxiang City, a pilot city for sponge city construction, are studied. Two sets of rainfall data and measured concentration data of runoff pollutants are used to determine the parameters of the Sarto-Boyd and p/r prediction models, and the evolution process of pollutant concentration in surface runoff are analyzed and simulated. The results show that the concentrations of surface runoff pollutants (SS and TP) in Pingxiang City are affected by the benefit of initial scouring and their peak values in the initial rainfall period reach the maximum with the peak of rainfall intensity, whereas the concentrations of the pollutants decrease obviously in the later period of rainfall, which are unlikely to be affected by the peak fluctuation of rainfall intensity. Restricted by the model parameters, the statistical indicators such as RSD, R2 and NSE of the Sarto-Boyd model are all lower than those of the p/r model, and the p/r model has better prediction accuracy for the surface runoff pollutant discharge law in Pingxiang City, so this model is more feasible. The p/r mathematical model established in combination with the rainfall situation of Pingxiang City can provide some reference for the prediction and management of pollutant concentration in the surface runoff during the construction of sponge city in the region.