The effects of the artificial bias on free air temperature trend derived from historical radiosonde data in China were assessed, by conducting quality control, homogenization and analysis on miss rate to the individual radiosonde temperature time series (RTTS) during 1958－2005. Identification reveals evident discontinuities in the RTTS with 2－4 breakpoints at 72% of stations. As a result of adjustments, original RTTS exhibit systematic cooling biases in the lower stratosphere and warming in the troposphere, suggesting both cooling in the stratosphere and warming in the troposphere are substantially overestimated in the original radiosonde data, e.g. the cooling trend at 100 hPa decreases 0.12 K/10 a and the warming trend at 700 hPa decreases 0.04 K/10 a. Missing rate is an important index to evaluate availability of RTTS for assessing regional average trend, and analyses suggest that the maximum missing rate of 30% is the optimum requirement for RTTS in China. Reduction of sampling stations results in underestimate of average trends in both the troposphere and the lower stratosphere. Upper air temperature changes in China are generally consistent with the results from the globe and the Northern Hemisphere, but there are some discrepancies: the atmosphere tends to warm below 500 hPa and cool in the upper levels with the maximum cooling around 300 hPa. Furthermore, the comparison of the trends among different sub-climatic regions implies spatial heterogeneity in upper air temperature changes in China.