Data-driven hydrological model can realize the accurate prediction for daily runoff in small watershed with less data types without considering the complex physical processes.Based on daily runoff monitoring data from Yuetan hydrological monitoring sites , in Huangshan of Anhui Province from 2009 to 2012,the paper built particle swarm optimization algorithm improved neural network (PSO-BPNN) and support vector machine (PSO-SVM) model.By comparing the results from different types of model, it discovered that both of the two models have good fitted ability and generalization ability, and PSO-SVM model based on three-day runoff data has the best simulation results and it can be used to predict daily runoff of Yuetan Basin and realize the rational allocation of water resources in the basin as well as the early prevention of related disasters.