The comparison and analysis are conducted by means of AWS (Automatic Weather Station) and manually observed pressure data from six stations in Gansu, so to identify the change resulted from the renewal of observation instruments and assess objectively the continuity and accuracy of observation data. The results indicate that AWS and manual records are very close in annual and monthly averages; the probability distributions of errors in regular, average, maximum, and minimum records are basically identical and relatively concentrated. The statistical test indicates that there is no significant differences in regular, maximum, minimum, 24record mean, and 4record mean measurements; there is obvious daily variation in regular records with the greater errors in 09:00〖CD*2〗10:00 and 19:00〖CD*2〗22:00 and smaller in 12:00〖CD*2〗15:00 and 01:00〖CD*2〗05:00; and there exist obvious differences among the six stations, in one or two of which there are manifest systematic errors due to the properties of observation instruments.