The global atmospheric temperature and humidity profiles are key datasets for studying extreme weather and climate change. However, the atmospheric profiles derived from satellite measurements include multiple error sources, such as instrument error, errors of retrieval algorithms, and quality control. Therefore, evaluating the errors of atmospheric profiles accurately is beneficial to understanding the features of derived atmospheric profiles and applying them in operations and research. The Three-cornered Hat (3CH) method, using the error model of X_i=truth+b+e_i, estimates the errors of various datasets by solving linear equations. Unlike the single data evaluation method, the 3CH method uses uncorrelated datasets and does not assume that any reference dataset is error-free. In this study, the errors of atmospheric temperature and humidity profiles derived from the Fengyun-3D polar-orbiting satellite (FY-3D), the Suomi National Polar-Orbiting Partnership (NPP), and the Constellation Observing System for Meteorology, Ionosphere and Climate 2 (COSMIC2) in 2021 are evaluated by the 3CH method. The differences between the 3CH method and the single data evaluation method, using ECMWF Reanalysis v5 (ERA5) as the reference dataset, are also discussed. The results show that: (1) The errors of three temperature and humidity profiles present similar vertical distributions. Especially around 700-300 hPa, the errors of all temperature profiles are less than 2 K, illustrating all temperature profiles are consistent and accurate in the middle and lower troposphere. In comparison of three humidity profiles, the FY-3D and NPP profiles present the largest and smallest errors, respectively. (2) Compared with all-sky situations, the errors of temperature and humidity profiles derived from FY-3D decline obviously in clear-sky situations. The differences in NPP profile errors between all-sky and clear-sky are very small, illustrating the effects of clouds on NPP profiles are minute. The COSMIC2 profile errors become a little larger at lower layers, because the occultation observation is affected by the super-refraction phenomenon over the ocean. (3) The errors estimated by the 3CH method are different from the results estimated by the ERA5 dataset, because the error of the ERA5 dataset and the correlations between three atmospheric profiles and the ERA5 dataset are included in the 3CH method. For the COSMIC2, the errors of temperature and humidity profiles estimated by the 3CH method increase by about 0.4 K and 0.3 g kg^-1 in the lower troposphere, respectively. In contrast, the errors of FY-3D and NPP profiles estimated by the 3CH method become smaller, which are more accurate than the results estimated by the ERA5 dataset.