Are “realistic” model atmospheres realistic enough?

During the latest decades the number of papers on stellar chemical abundances has increased dramatically. This is basically reflecting the very great achievements in telescope- and spectrometer-construction technology. The analysis of the resulting stellar spectra, however, is still not up to the standard that is offered by the observational methods. Recent significant advances in the analysis methods (i.e., in constructing model atmospheres and model spectra to compare with the observed ones) is reviewed with the emphasis on the application to abundance analysis of late-type stars. It is found that the very considerable progress that have been made beyond mixing-length convection and LTE is a major break-through for physically consistent modeling. Still, however, further steps must be taken, in particular for the cooler stars, before the situation is fully satisifactory.