An expression is obtained for the observed transmission, as measured by the sample in-sample out method, in terms of the transmission function of the sample, the spectral intensity distribution of the spectrometer and the atmospheric transmission function A(ν). The effects of atmospheric absorption can be eliminated only if A(ν) is constant; the nature of the systematic errors arising when A(ν) is not constant is indicated by means of examples. It is concluded that no simple corrections are possible and the only certain way to eliminate these effects is to carry out the measurements in a vacuum.