Abstract: Radiosonde observations are crucial in weather forecast and upper-air climate research. Due to their high vertical resolution, they are also important for calibration and validation of satellite temperature and water vapor retrievals. Quality control of Chinese radiosonde observations of 1951-2012 is conducted with hydrostatic check, and the result reveals that ratios of erroneous geopotential height (GPH) of 900 hPa, 800 hPa, 80 hPa and 60 hPa before 1963 are 2%-11%, which are much higher than other levels. The ratios of erroneous GPH of 250 hPa and 70 hPa in mid of the 1970s of a few stations in Yunnan Province are also much higher than other levels. Changes with observation code and practice in Chinese radiosonde observations are thoroughly investigated. Then the possible cause of the phenomenon is analyzed and validated with observations.The height of each pressure level in sounding is calculated by accumulating the thickness of layers between each pressure levels below on the base of station elevation. And the thickness is determined by hydrostatic equation automatically nowadays, but due to the limitation of calculation capacity before the 1980s, they are available in ready-made GPH-table. However, 900 hPa, 800 hPa, 80 hPa and 60 hPa are not included in the GPH-table possibly because they are not directly used in weather charts then. So GPHs of levels in question are obtained through interpolation on time-GPH chart after GPH of other levels are available on the chart through looking up the GPH-table. This practice makes GPH of levels in question vulnerable to human mistakes, thus many records contain errors. The explanation to the cause of the problem is validated with observations.Furthermore, the method to correct the erroneous GPH record is proposed, which is to recalculate it based on hydrostatic equation with correct observations of other levels from the same sounding. The correcting method is validated since the originally correct GPH records are reproduced accurately. For 900 hPa and 800 hPa of Ganzhou Station, differences between reproduced GPHs and original ones are less than 5 gpm, and about 50% of reproduced GPHs are exactly the same as original ones. Differences for 250 hPa of Kunming Station are all less than 5 gpm either, but differences are slightly larger for 70 hPa, possibly because the layer between 70-100 hPa is as large as 2300 gpm approximately, and significant levels are not enough to reveal all the details of temperature profile. Besides, the correcting method is validated from another aspect, since after the erroneous GPH are corrected, the relevant hydrostatic residue series become consistent.