Many impact models (e.g., hydrological and agricultural models) require simulations of weather variables reflecting the spatial and temporal dependence of observed meteorological fields. New techniques are recently available to generate weather variables simultaneously at multiple locations. This paper presents a comparison of two types of multi-site stochastic weather generators (MulGETS model and k-NN model) for simulation of precipitation and temperature at a network of 12 stations in Xiang River Basin, China. These two models were evaluated for their ability to reproduce the statistical features of the historical meteorological field. The results showed that both MulGETS and k-NN model were successful in reproducing the mean, standard deviation, and skewness of the weather variables, while the performance of k-NN was generally superior to that of MulGETS. The k-NN model was found to perform satisfactorily in preserving the spatial structure of the weather variable, especially the spatial intermittence. Only MulGETS model could generate extreme values out of the historical range. New technology is needed because both MulGETS and k-NN model have the limitation in representing temporal dependence of weather sequence, especially the autocorrelation of daily precipitation.