In arid regions, groundwater is recharged laterally from rivers; this recharge is the primary source of water for vegetation. The rise or decline of the water table results in salinization or desertification respectively, and damages vegetation growth. Therefore, prediction of the water table with consideration of the stream-aquifer interactions is very important for understanding the hyporheic processes under human intervention, and to assess the ecological effects and water resource management of arid regions. This paper used a statistical-dynamical approach to predict the water table elevation near a river channel in an arid region from river discharge, which was achieved by a numerical model, a fitted stage-discharge function, and an automatic parameter calibration method. The model was established by reducing the problem of water table prediction under stream-aquifer interactions to an initial value problem of the differential equations with prescribed river discharge based on Darcy's law; the model parameters were automatically calibrated by the Shuffled Complex Evolution method. Synthetic experiments were used to test the sensitivities of the water table prediction to model parameters, and showed the robustness of the proposed scheme under different conditions. The proposed scheme was also calibrated and validated using data collected at the Yingsu section in the lower reaches of the Tarim River. The systematic biases, mean absolute errors, root mean squared errors, and correlation coefficients for the validation had ranges of -0.22 similar to 0.16 m, 0.14 similar to 0.33 m, 0.18 similar to 0.44 m and 0.94 similar to 0.99 m respectively. Copyright (C) 2009 John Wiley & Sons, Ltd.