The evolutions of Asian summer monsoon (ASM) are detected and evaluated based on the models in Couple Model Intercomparison Projects Phase-3 and Phase-5 (CMIP3 and CMIP5) for the 20 th Century climate simulation (20c3m and Historical runs, respectively). Considering that the individual models have various biases in rainfall amount simulation, instead of applying a fixed rainfall criterion as used in observation, we use model-dependent rainfall criteria to identify the simulated ASM onset, retreat, and duration. This model-dependent criterion is defined as the height in cumulative distribution function (CDF) of simulated precipitation that the observed criterion occurs. Based on this method, the multi-model ensembles (MMEs) of CMIP3 and CMIP5 both show a delayed monsoon onset but an earlier retreat relative to the observations, indicating that models tend to underestimate the monsoon period. The MME results show a skill in capturing the ASM domain which features monsoon rainfall characteristics, whereas a large spread is found among individual models. Overall, the state-of-the-art CMIP5 models show slightly improvements from the CMIP3 models in the simulations of ASM domain and evolutions. Models with a hybrid method based on bulk mass flux and CAPE closure schemes perform better than models with other types of convection parameterization.(Citation: Tung, Y., C. Chen, and P. Hsu, 2014: Evolutions of Asian summer monsoon in the CMIP3 and CMIP5 models. SOLA, 10, 88−92,