Simulations of a regional (approx. 50 km resolution) circulation model REMOiso with embedded stable water isotope module covering the period 1958-2001 are compared with the two instrumental climate and four isotope series (d18O) from western Svalbard. We examine the data from ice cores drilled on Svalbard ice caps in 1997 (Lomonosovfonna, 1250 m asl) and 2005 (Holtedahlfonna, 1150 m asl) and the GNIP series from Ny-angstrom lesund and Isfjord Radio. The surface air temperature (SAT) and precipitation data from Longyearbyen and Ny-angstrom lesund are used to assess the skill of the model in reproducing the local climate. The model successfully captures the climate variations on the daily to multidecadal times scales although it tends to systematically underestimate the winter SAT. Analysis suggests that REMOiso performs better at simulating isotope compositions of precipitation in the winter than summer. The simulated and measured Holtedahlfonna d18O series agree reasonably well, whereas no significant correlation has been observed between the modelled and measured Lomonosovfonna ice core isotopic series. It is shown that sporadic nature as well as variability in the amount inherent in precipitation process potentially limits the accuracy of the past SAT reconstruction from the ice core data. This effect in the study area is, however, diminished by the role of other factors controlling d18O in precipitation, most likely sea ice extent, which is directly related with the SAT anomalies.