Current national ground meteorological observation net consists of the benchmark, basic and general stations (counting up to 2416 stations), of which about 65.8% the general stations didn’t observe at night time. According to Specification of Surface Meteorological Observation， the average daily temperature must be calculated through mathematical methods, giving a replacement value for 02 BT’s temperature. This process virtually brings about calculation errors. Using the timing temperature data of 19 national basic stations in Shaanxi from 1961 to 2009, the assignment errors are analyzed and evaluated quantitatively through various schemes comparison, and a new assignment (or interpolation) method is raised. The results show that: (1) The replacement value of 02 BT’s given by traditional method is higher than the observed temperature, the average difference at 19 stations is 0.77℃. (2) Radiation intensity caused by stations in place of latitude, season and weather conditions conducting to assigned variable (20 BT’s temperature the day before) magnifies obviously 02 BT’s temperature assignment error. (3) Using stepwise regression analysis method to adjust the traditional weighted average as multifactor asymmetric weighting method can reduce significantly the weighting coefficient of 20 BT’s temperature the day before and reduce the influence caused by the intensity of radiation, thus making the assignment closer to the observed one. The average difference at 19 stations is approximately zero. New regression equations have certain correction effect on the monthly and yearly average temperature, and the annual average temperature can be decreased by 0.1 to 0.4℃, the monthly average temperature even by 0.5℃.