The accuracy of Automatic Weather Station (AWS) temperature data can be affected directly by the processing performance of the data acquisition unit for temperature channel. The Least Squares Support Vector Machine (LSSVM) method is used to establish the model with the data from the temperature calibration experiment on the data acquisition unit for temperature channel in an unattended AWS to correct measurement errors. The measurement uncertainty of the corrected temperature channel is evaluated. Results show that after correction, the measurement error of the data acquisition unit for temperature channel is less than 005 ℃ and the uncertainty is only 006 ℃, which are far less than the values before correction. This method is applicable in other meteorological data acquisition devices to guarantee the quality of AWS observation data.