This study is stimulated by the finding that there is a decreasing trend of retrieved cloud droplet size during the lifetime of NOAA-9 and a sudden change of droplet size if different satellite data for the same period were used in the retrieval. This trend and sudden change in the retrieved results are apparently artifact rather than natural variations. Since channel 3 data of the AVHRR are used in the retrieval, the calibration method is examined to explore the possible explanation. Until now, no effort has been made for radiance calibrations of channel 3 although many investigations have been conducted for the visible and thermal infrared channels of AVHRR. There are two reasons for this lack of activity. First, detectors used for channel 3 of AVHRR are different from those for channels 4 and 5 and should not be subject to non-linearity. Second, the channel 3 of AVHRR has in-orbit calibration by looking at an internal calibration target (ICT) and space. This in-orbit calibration procedure is appropriate for earth temperature monitoring because the ICT radiates at around the earth surface temperature, and the intensity peaks at about 10 micrometer. There are evidences showing that the channel 3 brightness temperature does not change with satellites. However, in some other applications such as the retrievals of cloud particle sizes, the calculation of reflectance depends on the solar energy that radiates at much higher temperature (approximately 5800 K); the intensity of radiated solar energy peaks at about 0.50 micrometer. Therefore, the spectral response change of the instrument may lead to different values of in-band solar flux and thus cause errors in calculations of the reflection function of solar energy. For example, if the sensitivity of the response curve decrease is less in the short-wavelength end than at the long-wavelength end within the band after launch, then the in-band solar energy will be larger even if the channel 3 brightness temperature is calibrated against to the ICT. This will cause an overestimate of solar reflection function and an underestimation of droplet size because pre- launch value of in-band solar flux is used in the calculation. Several periods of satellite data have been used for this investigation: the overlapped AVHRR observations during January, 1985 for three weeks by NOAA-7 and NOAA-9; one month of NOAA-9 data of Oct. 1987 and Oct. 1989. The results of the study show that the ICT works well for calibration of thermal emission from the Earth's subjects. However, there are evidences showing that the in-band solar energy of channel 3 of AVHRR is underestimated for NOAA-7 comparing with NOAA-9 and this constant is increasingly underestimated during the satellite life-time of NOAA-9.