In this talk I presented and discussed our unexpected detection of water vapor in the disk-averaged spectrum of the K2IIIp red giant Arcturus [for details, see Ryde et al. (2002)]. Arcturus, or alpha Bootes is, with its effective temperature of 4300 K, the hottest star yet to show water vapor features. We argue that the water vapor is photospheric and that its detection provides us with new insights into the outer parts of the photosphere. We are not able to model the vater vapor with a standard, one-component, 1D, radiative-equilibrium, LTE model photosphere, which probably means we are lacking essential physics in such models. However, we are able to model several OH lines of different excitation and the water-vapor lines satisfactorily after lowering the temperature structure of the very outer parts of the photosphere at log tau_500=-3.8 and beyond compared to a flux-constant, hydrostatic, standard marcs model photosphere. Our new semi-empirical model is consistently calculated from the given temperature structure. I will discuss some possible reasons for a temperature decrease in the outer-most parts of the photosphere and the assumed break-down of the assumptions made in classical model-atmosphere codes. In order to understand the outer photospheres of these objects properly, we will, most likely, need 3D hydrodynamical models of red giants also taking into account full non-LTE and including time-dependent effects of, for example, acoustic wave heating sensitive to thermal instabilities.