In AR4 chapter 2, page 153, the IPCC states that 

"Harries et al. (2001) analysed spectra of the outgoing longwave radiation as measured by two satellites in 1970 and
1997 over the tropical Pacific Ocean. The reduced brightness temperature observed in the spectral regions of many of the
greenhouse gases is experimental evidence for an increase in the Earth’s greenhouse effect. In particular, the spectral signatures were large for CO2 and CH4." 

The 2001 paper by Harries et al (on-line here) made use of data from the IRIS satellite, launched in 1970, and a comparison with data from the much more recent (1996) IMG satellite.  They estimated errors in the results of around 0.5K.

However, a more recent paper by Brindley and Harries  in the Journal of climate in 2003 (here) raises some doubts about these results, saying that "the IMG sampling is too sparse and yields results that differ from the true case by up to 6.0 K".  They also make the cautionary statement that "Comparisons with the observed IMG–IRIS difference spectra show that these uncertainties due to sampling presently limit the conclusions that can be drawn about climatically significant feedback processes".

This more recent paper, by the same authors, saying that the errors were much larger than previously thought and urging caution in the interpretation, is not cited by the IPCC.  

This is yet another example of the IPCC misleading its readers by cherry-picking from the literature, ignoring research that does not fit the IPCC story. 

For more on this, see the Niche Modeling blog.