Abstract: Based on surface observation data archived by the National Meteorological Information Centre, a dataset of the 1991–2020 China climate normals is developed. For the development of the dataset, the metadata of the observation data since 1991 is systematically checked, verified, and revised. All elements are segmented using the station location migration information, and the segmented processing of the temperature series is completed based on the heterogeneity test results. The segmented processing of visibility observation data is completed using manual and automatic observation and adjustment information. The segmented processing of visibility observation data is completed, providing a better representative climate background field consistent with the current visibility observation method. The Fourier series theory is used to deal with the harmonics of the annual daily value series of temperature and precipitation, reflects the seasonal transformation of meteorological variables, and avoids abnormal mutation characteristics. The 1991–2020 China climate normals provide the climate background information of 14 elements, i.e., temperature, precipitation, air pressure, wind direction and speed, relative humidity, water vapor pressure, cloud, weather phenomenon, visibility, evaporation, snow, ground temperature, frozen soil, and sunshine, of 2438 stations in China and data support for weather and climate research.