Molecular modelling is of major help to understand structure-function data on G-protein-coupled receptors (GPCRs). Since the first determination of the structure of rhodopsin, at high resolution, the view has emerged that it will be now easy to automatically obtain realistic models for any GPCR by homology modeling. Our experience on cholecystokinin CCK(1) receptor modelling together with available data on other GPCRs leads us to rule out this opinion. We believe that construction of realistic models of certain GPCRs still remains time-consuming and requires many refinements of the models in close association with experiments. This conclusion has important consequences for modelling orphan GPCRs.