In the last 50 years, the ¹⁸O/¹⁶O signature of meteoric water (in the following, ^σ18O<inf>prec</inf>) has become a key tracer intensively used both in hydrology and in palaeoclimatology. In palaeoclimatology, the original atmospheric signal, i.e. ^σ18O<inf>prec</inf>, is reconstructed by means of a variety of palaeo-archives such as ice cores, l acustrine sediments, tree ring cellulose or speleothems. In both research fields, the understanding of the original isotope signal, before entering a hydrologic system or b efore being archived, is essential. To this end, a powerful modelling tool was developed about 20 years ago: Atmospheric General Circulation Models (AGCMs) fitted with water isotope diagnostics (Hoffmann et al., 1998; Joussaume et al., 1984; Jouzel et al., 1987). The fitting procedure is a relatively straightforward, although technically difficult process. The hydrologic cycle as described in these three-dimensional AGCMs has to be replicated, and each time a phase transition is calculated in the model the corresponding fractionation process has to be taken into account to estimate the isotopic composition of the various water fractions. In this way, the water isotopes are computed solely depending on climate, with their variability simulated by the AGCM. The AGCM itself is constrained, depending on the time scale considered, by observed or reconstructed sea surface temperatures (SSTs), atmospheric greenhouse gas concentrations and solar insolation. To date, at least half a dozen modelling groups have followed this approach (see Table 1) and additional modelling groups are expected to build such an isotope module into their GCMs. The focus of interest of the various scientists using the results of three-dimensional water isotope models is different, according to their main working field and speciality. GCM modellers, for example, are interested in an independent test of the parameterization of the model's water cycle. A variety of basically unobserved variables such as condensation temperatures, cloud thickness, evaporation rates, etc. influence the σ¹⁸O<inf>prec</inf> signal. Therefore, comparison of modelled and observed water isotope signals presents a valuable test for this important part of the physics of AGCMs. Hydrologists, on the other hand, are increasingly applying their models to larger spatial scales. This typically involves the need to apply new and uncertain parameterization techniques and again calls for an independent and sensitive test procedure, as provided by water isotopes. Spatial scales of AGCMs and some hydrologic models are now sufficiently close to each other that a physical coupling will be feasible soon. Finally, palaeoclimatologists certainly have the most vivid interest in a water isotope modelling tool that allows the interpretation of σ¹⁸O<inf>prec</inf> not only in terms of local temperatures or precipitation rates but also in terms of atmospheric circulation and source region changes, for example. Classically, palaeoclimatologists do the necessary calibration of an isotope palaeo-record using a modern analogue technique. They establish a relationship between regional σ¹⁸O<inf>prec</inf> and climate parameter (for example annual Tsurface) and subsequently apply this modern spatial relationship to the temporal σ¹⁸O<inf>prec</inf> series they have measured. However, it is well known that such modern analogue methods involve many uncertainties. The most important example of its failure is probably the application of the modern spatial σ¹⁸O<inf>prec</inf>/ temperature relationship in interpreting the ice core records in Central Greenland. Here, a number of independent studies have shown that, at least on a glacial/interglacial time-scale, the real temporal σ¹⁸O <inf>prec</inf>/temperature slope is about half the modern spatial slope (Cuffey et al., 1995; Dahl-Jensen and Johnsen, 1986; Severinghaus et al., 1998). It is certainly one of the major successes of "isotopic AGCMs" that they were able to give a sound explanation for this de-calibration of the isotopic thermometer by simulating a strong shift in the seasonal distribution of precipitation under full glacial conditions (Werner et al., 2000c). This Paper reports on advances in modelling water isotopes by means of GCMs during the last five years. After briefly summarizing the fractionation physics built into AGCMs (Section 2), we present results of the ECHAM "isotopic AGCM" integrated under boundary conditions corresponding to the 20th century. In Section 3, we present a model/data comparison for the most recent version of the ECHAM AGCM, the cycle 4 version of the climate model of the Max-Planck Institut für Meteorologie, Hamburg. In an earlier paper (Hoffmann et al., 1998), results of the cycle 3 version of the ECHAM model were compared with observations of the IAEA/GNIP network. Here, we did a systematic model evaluation of the ECHAM 4 version in T30 resolution on a seasonal to interannual scale. The model/data comparison presented here is designed to include "palaeo-isotope" series in the future and to extend this kind of study to the interdecadal and centennial time-scale. © 2005 IEA.