Abstract: The gridded observational air temperature dataset (CN05.1) for the period 1986-2005 over China is used to evaluate daily extreme temperature indices simulated by 21 models that participate the NASA Earth Exchange/Global Daily Downscaled Projections (NEX-GDDP). Four extreme temperature indices, including the lowest daily temperature maximum (TNx), the highest daily temperature (TXx), the warm night frequency (TN90p) and warm day frequency (TX90p), are adopted to investigate the change of extreme temperature. The major conclusions are as follows. (1) Except for TXx from the MRI-CGCM3, the four indices from other models show an upward tendency, which is consistent with observations. However, the magnitudes of their linear trends are less than that from observations with the values of 0.26℃/decade (TNx), 0.19℃/decade (TXx), 2.21% decade (TN90p), 1.04%/decade (TX90p), respectively. (2) There are large differences in spatial patterns of those indices between models. For the simulation of all the four indices, CCSM4 performs the best, CESM1-BGC, MIROC-ESM-CHEM ranking next in order of performance. The spatial patterns of climatological extreme indices can be simulated perfectly with the correlation coefficients of observations with TNx and TXx from all models exceeding 0.97. The ratios of standard deviations between simulations and observations for TN90p and TX90p vary from 0.34 to 1.58, and the root mean square errors are within 1.6%-3.47%. (3) Synthetical evaluation of the four extreme indices in term of their climatological means and linear trends indicates that the performances of three models (i.e., CanESM2, CESM1-BGC and MIROC-ESM-CHEM) are relatively better. Therefore, it is suggested that results of the above three models in the NEX-GDDP can be used to investigate the extreme temperature change in the future.