This article presents an analysis of the spatiotemporal changes (1960–2100) in temperature and precipitation extremes of a typical arid zone (i.e., the Tarim River Basin) in Central Asia. The latest observations in the past five decades (1960–2009) and Coupled General Circulation Model (CGCM) multimodel ensemble projections (2010–2100) using the Bayesian Model Average (BMA) approach are employed for analysis in this study. Results indicate: (1) Most warm (cold) extreme temperature indices have shown significantly positive (negative) trends in the Tarim River Basin in past five decades, while only slight changes in precipitation extremes can be observed. From the spatial perspective, more significantly warm (cold) extremes are found in the desert zones than in upstream mountain zones (i.e., the Tian Shan Mountain and Kunlun Mountain systems which surround the basin). Whereas, there are no identical spatial patterns for the change in extreme precipitation; (2) Ensemble of five CGCM models in Phase 3 of the Coupled Model Intercomparison Project (CMIP3) based on the BMA method suggests that the increasing consecutive dry days (CDD), together with the decreasing frost day (FD) and increasing warm nights frequency (TN90) may lead to more frequent droughts in Tarim in future. Meanwhile, slight increase of annual count of days with precipitation of more than 10 mm (R10), maximum 5-day precipitation total (R5D), simple daily intensity index (SDII), and annual total precipitation with precipitation >95th percentile (R95) in projections indicate a probability of flood occurrence in summer together with frequent occurrence of droughts. The results can provide beneficial reference to water resource and eco-environment management strategies in arid zones for associated policymakers and stakeholders.