Based on the daily mean air temperature from 1971 to 2000 observed by more than 2000 national surface stations in China, a standardized method was employed to carry out missing data estimation experiment, and the results were verified by cross validation. Two schemes, the relation optimal scheme and the closest station scheme, which were both used to pick up the adjacent stations, were compared. It showed that the relation optimal scheme was better than the other, and only 4 adjacent stations that are most closely related to the estimated station were necessary for estimation. The results indicated that estimate values in average deviate from true values by 0.42℃. The absolute mean error and root mean square error between the estimation and the actural measurements and the sample ratio with the differences falling in ±0.5℃ were all shown good exponential relationships with the average adjacent station distance.