Abstract: Runoff prediction is helpful to the comprehensive and efficient allocation of water resources and flood control and disaster reduction operation in the basin. How to accurately carry out a short-term runoff prediction has always been the focus of hydrology and water resource research. Gaussian process regression（GPR） has been successfully applied in the long, medium and short-term hydrological process prediction research because of its generalization ability for complex nonlinear regression problems. The GPR regression analysis ability depends on not only the model parameters but also the kernel function. Therefore, this paper analyzes the effect of GPR prediction model under different kernel functions, and proposes a short-term runoff prediction model based on exponential kernel function. Through the multiple correlation coefficient analysis of the largest multiple correlation coefficient, and the shortest predictor period, and then the rational secondary, radial base, maton and exponential kernel function are chosen to establish different GPR short-term runoff prediction model, also joined the MLR, RT, SVM, BP model method prediction results as a comparison. Taking the short-term runoff prediction of Ji′an Hydrologic Station in the Ganjiang River Basin（the prediction step is 6 h, and the prediction period is 7 days） as an example, the relevant experimental results show that:（1） There are obvious differences in the prediction results of GPR models using different kernel functions, and the prediction performance of different methods from good to bad is exponential GPR, rational quadratic GPR, RT, Marton GPR, Radial basis GPR, SVM, MLR, BP；（2） The 4 evaluation indexes of the exponential GPR prediction model in 28 periods all performed best, DC and QR are close to 1and 100% respectively, the forecast accuracy reaches grade A or above. In conclusion, this paper verifies the effectiveness and universality of the exponential kernel function GPR short-term runoff prediction model, and the model prediction accuracy meets the needs of practical engineering applications with practical application value.