To predict daily runoff timely and accurately plays an important role in the reasonable planning,utilization and management of water resources.This paper built daily rainfall-runoff model to predict daily runoff for seven days in Pailugou catchment in a typical catchment of Qilian mountains based on support vector machine(SVM).In order to test the validity of the developed model,it compared the results between SVM model and traditional artificial neural network(ANN) model in terms of different evaluation criteria during validation period.Results showed that both SVM and ANN presents very high precision and SVM model performed better than ANN model.The SVM model may be considered as an effective tool to establish a medium and long-term daily runoff forecast model in semiarid mountain regions under limited data condition.