Abstract. Oscillations of the level of the Caspian Sea (CS) primarily stem from variations in runoff from the Volga River. Therefore, changes in the level of the CS can be used to assess the ability of climate models to reproduce the water budget over the East European Plain. We compare observed or reconstructed CS level positions during the Last Glacial Maximum (LGM), Holocene and modern periods with the CS level positions calculated based on simulations in experiments using the CMIP5/PMIP3 protocol. The proxy dataset comprises a number of detailed maps of the CS for the main regression stages and transgression stages during the last 30 ka together with information about sea-level positions. The results show poor correspondence between the model simulations of decadal- and centennial-scale level oscillations and proxy reconstructions. We propose that such discrepancies can be caused by insufficient sensitivity of the climate models used. The modelled data could be verified based on how well the models simulated the sharp decrease of the Volga River runoff during the LGM, which caused a large decline of the CS level.