<p>The Tarim River Basin (TRB), China’s largest endorheic basin, lacks comprehensive hydrological understanding, particularly due to its desert groundwater system. To fill this gap, a GSFLOW-based groundwater/surface water model was developed for plains/desert areas, coupled with mountain snowmelt routing models. These hydrological models were calibrated through Bayesian uncertainty analysis, and validated using streamflow and groundwater level observations, as well as remote sensing data. The validation results demonstrated that the calibrated models effectively characterize the snowmelt runoff and groundwater/surface water processes in the TRB. Streamflow from mountains and evapotranspiration are the primary inflow and outflow fluxes in the plain and desert areas of the TRB, accounting for 339.7 × 10⁸ and 335.1 × 10⁸ m³/year, respectively. Groundwater abstraction, particularly during irrigation periods, has led to a downward trend in groundwater storage, with an annual extraction volume reaching approximately 56.3 × 10⁸ m³/year. Notably, the study highlighted that the groundwater in the Taklamakan Desert is not an isolated system but engages in significant exchange with surrounding areas, with river leakage and lateral groundwater inflow serving as the primary recharge sources (1.11 × 10⁸ and 2.60 × 10⁸ m³/year). The primary discharges of groundwater in the desert include groundwater evaporation and lateral groundwater discharge, with rates of 5.11 × 10⁸ and 2.05 × 10⁸ m³/year, respectively. The total groundwater storage in the Taklamakan Desert exhibits a downward trend, with a rate of −1.85 × 10⁸ m³/year. This study addressed the knowledge gap on the characteristics of the groundwater system in the Taklamakan Desert and provides a comprehensive understanding of the complex hydrological processes in the TRB.</p>