Empirical distribution functions were applied for removing long-term errors from BT data derived from AVHRR sensor on NOAA environmental satellites. This paper investigates BT stability in the NOAA/NESDIS Global Vegetation Index (GVI) data set during 1982-2003. This period includes five NOAA satellites. Degradation of BT over time for each satellite was estimated for geographical location in China. The method of matching empirical distribution function (EDF) improves the time relative stability of BT data for all satellites, especially NOAA-9, -11 and -14.