The compositing technique used to derive global vegetation index (NDVI) from the NOAA-AVHRR radiances, reduces the residual effect of water vapor and aerosol on the NDVI. The reduction in the atmospheric effect is shown using a comprehensive measured data set for desert conditions, and a simulation for grass with continental aerosol. A statistical analysis of the probability of occurrence of aerosol optical thickness and precipitable water vapor measured in different climatic regimes is used for this simulation. It is concluded that for a long compositing period (e.g. 27 day), the residual aerosol optical thickness and precipitable water vapor is usually too small to be corrected for. For a 9 day compositing the residual average aerosol effect may be about twice the correction uncertainty. For Landsat-TM or EOS-MODIS data, the newly defined atmospherically resistant vegetation index (ARVI) is more promising than possible direct atmospheric correction schemes, except for heavy desert dust conditions.