Abstract: In the next several years, the conventional manual surface observational systemused for the past several decades in China will be completely replaced by automatic weather stations (AWS). It is inevitable that there are systematic differences between observations recorded by different systems (including different sensors and different screens). In climatological viewpoint, it is very important to check the differences of temperature records obtained by the two observation systems and to find the causes responsible for the differences. This study was undertaken by using two kinds of comparative observations, one is the hourly parallel observation of temperature, cloud cover, and precipitation recorded by AWS (the Milos500) and manual observation from 2001 to 2003 to analyze the difference of temperatures due to the updating of instrument, and other is the parallel observation using traditional wooden screen and glass fiber reinforced plastic screen to analyze the influence of the updating of screen on temperature observations. Based on the parallel observation data obtained at the same meteorological platform by the AWS and manual observation systems (i.e. by different sensors in the same type screen) at 5 reference climatological stations in 2001-2003, the differences of hourly, daily and monthly temperature observations and temperature statistics were analyzed, and the causes of the differences are interpreted. We have quantitatively estimated the deviations resulted from the accuracy and sensitivity of the AWS and conventional observation instruments, and solar/infrared radiation. The updating of observation instrument evidently impacted the daily, monthly, and annually mean temperatures, leading to the temperature difference of about 0.2 ℃, and the effect of solar radiation on the two kind of observational instruments is the major reason. Besides, two instruments had a systematic error about 0.1 ℃, and difference in the sensitivity of the two instruments to the environmental temperature change also brought about a temperature difference of 0.1-0.15 ℃ in different time periods in a day. The analyses of the comparative observation data of temperatures recorded respectively in traditional wooden screen and glass fiber reinforced plastic screen at 3 stations indicate that impact of the updating of screen on temperature observations was minor, but the observed value of temperature in glass fiber reinforced plastic screen was more sensitive to changes in the environmental temperature.