Abstract: The consistency of upper-air observations in serial of space and time is attached more and more importance especially by the users and upper-air observation operators. Generally, the systematic differences of different types of radiosondes can be obtained using the method of direct comparison when the compared radiosondes are suspended below the same balloon. However, as these errors are more or less related with the releasing place, time and method, the results of a few direct comparisons cannot reflect the differences properly.It is found that FG (first guess) fields can reflect the regional differences of meteorological element arose by weather process generally and such differences can be removed indirectly using FG. If two different types of radiosonde systems are equipped at neighborhood stations for quite a long time, they can be compared indirectly using FG.Average bias is analyzed between the observations and FG from neighborhood stations of Urumqi, Yining, Klamayi, Beitashan (4 stations which use P-band radiosonde system), Altay, Aksu and Kuerle (3 stations which use L-band radiosonde system) during 2007—2010. The result reveals that the differences of average bias between observations and FG from L-band stations are small during both daytime and night. The differences of average bias at the same sounding time from the 4 P-band stations are normal, but the differences are very large between 0800 BT and 2000 BT.And the differences of average bias are very small between P-band and L-band at 2000 BT but large at 0800 BT.It shows that observations at 0800 BT by P-band radiosonde system might be problematical, which call for more attention when using these data in Xinjiang. Statistical results are also given as reference for correcting the systematic error.