In infrared remote sensing by Fourier transform spectroscopy usually not the measuredrnspectrum as a whole is analysed hut subsets, so-called "microwindows", whichrncontain prominent transitions of the target species. While microwindows noted inrnthe literature were selected by conventional methods, the purpose of this work isrnto select optimum microwindows for analysis of target species such that the relatedrnretrieval error is minimized. The choice of microwindows is shown to depend onrnthe assumed retrieval scenario. In particular microwindow boundaries and associatedrnretrieval errors are sensitive to assumptions on errors of parameters which arernnot jointly fitted along with the target parameters but kept constant during thernretrieval. It is shown that consideration of uncertainities in pressure, temperaturernand measurement geometry has major influence on the optimization of microwindowrnboundaries and the selection of microwindows. Some of the microwindows whichrnare optimum in the idealized case of negligible uncertainties of pressure, temperaturernand measurement geometry, are similar to those reported in the li terature,rnwhile there are considerable differences between the latter and those microwindowsrnoptimized for more realistic conditions. Hence our anal ysis suggests that existingrnmicrowindows defined by conventional methods are not best suitable for applicationrnto realistic measurement scenarios because these sources of systematic errors havernbeen underestimated