Abstract: In order to solve the problem of missing record of temperature observation, taking the Hubei Province in 2020 as an example, we selected three methods, that is, Inverse Distance Weighted (IDW), Ordinary Kriging (OK) and Multiple Linear Regression (MLR), to interpolate the missing values of daily mean temperature (T), maximum temperature (T_max), and minimum temperature (T_min). Based on this interpolation results, using the average absolute error (MAE), we evaluated the interpolation results obtained by these three methods. The results show that the MAE of T_max, T_min and T obtained with the MLR interpolation is the lowest, which are 0.41℃, 0.31℃, and 0.20℃, respectively. Meanwhile, the interpolation errors of T at all stations are less than 1℃. Compared with IDW and OK, the MAE spatial distribution of interpolation results obtained with MLR is more uniform with slight changes with altitude and seasons. Single station test shows that the more samples used for MLR model is and the greater the sample time dispersion is, the better the interpolation effect of MLR on temperature is. On the whole, the interpolation effect of MLR on missing values of daily temperature from regional stations is the best, IDW is the second, and OK is the worst. For establishing long-term continuous temperature datasets of meteorological stations, MLR is more suitable for solving the problem of missing records of daily temperature from regional automatic weather station (AWS).