Using radiosonde temperatures of 92 selected stations in China,the uncertainties in homogenization processes caused by different reference series,including nighttime temperature,the NCEP (National Centers for Environmental Prediction) and ERA-40 (European Centre for Medium-Range Weather Forecasts) forecasting background,are examined via a two-phase regression approach.Although the results showed limited consistency in the temporal and spatial distribution of identified break points (BPs) in the context of metadata events of instrument model change and correction method,significant uncertainties still existed in BP identification,adjustment,and impact on the estimated trend.Reanalysis reference series generally led to more BP identification in homogenization.However,those differences were parts of global climatic shifts,which may have confused the BP calculations.Discontinuities also existed in the reanalysis series due to changes in the satellite input.The adjustment values deduced from the reanalysis series ranged widely and were larger than those from the nighttime series and,therefore,impacted the estimated temperature trend.