In this study, we use observational datasets to evaluate the performance of 34 models participating in the Coupled Model Intercomparison Project Phase 5 (CMIP5) in simulating dry- and wet-season precipitation over southwestern China during 1986-2005. Of the 34 CMIP5 models, 30 and 25 models overestimate dry- and wet-season precipitation, respectively. The ability of 34 models to simulate dry- and wet-season precipitation is found to differ significantly. Moreover, approximately half of the models show that spatial correlation coefficients with the observations are significant at the 99% confidence level and the ratios of the simulated standard deviations to the observed values are less than 2. On the basis of the two skill scores, we select the nine best models for simulating dry- and wet-season precipitation. The 9-model ensemble mean performs better than the ensemble mean of all 34 models and most of the individual models. Therefore, we further use nine-model ensemble mean to project dry- and wet-season precipitation changes under Representative Concentration Pathways RCP4.5 and RCP8.5 scenarios over southwestern China. Compared with the climatology of 1986-2005, the dry-season mean precipitation is projected to increase in the west Sichuan plateau and decrease in the Sichuan Basin, the Panxi region, Chongqing, Guizhou, and the most of Yunnan province during 2016-2035. During the same period, the wet-season mean precipitation is projected to increase in west Sichuan plateau and most of Guizhou and Guangxi and decrease over Sichuan basin, the Panxi region, and Yunnan Province. Both dry- and wet-season precipitation is projected to increase throughout nearly the entire southwestern China region under the two scenarios in the middle and late 21st century. Under the RCP8.5 scenario, the changes in magnitude of dry- and wet-season precipitation are stronger than those under the RCP4.5 scenario.