We used the ERA5 data from European Centre for Medium-Range Weather Forecasts to evaluate the accuracy of regional weighted mean temperature (T_m) linear models over China.These models were derived from historical radiosonde data of 88 Chinese stations.For these stations,the root mean square errors (RMSEs) between single-station model derived T_m and ERA T_m are 1.8-5.5 K,and -1.22-4.54 K of which are system biases.For 82 out of the total 88 stations,the system biases are positive,indicating that the model derived values are generally larger than ERA T_m.The RMSEs can be reduced to 1.5-3.5 K after removal of the system biases.On average,the RMSE of T_m from single-station model is smaller than that from the unified model by 0.6 K,suggesting that single-station models are generally more accurate than the unified model.In western China,northwestern China and Inner Mongolia,the accuracy can be improved by 1-3.9 K using single station models compared with the unified model.Time series of the biases between model derived T_m and ERA T_m show obvious seasonal cycles for over half of the stations.