Based on the remote sensing data from National Snow and ICE Data Center (NSIDC), the performance of CMIP5 (Coupled Model Inter-comparison Project) models in reproducing the winter snow water equivalent (SWE) in the Eurasian continent during 1981-2005 was evaluated first, and the multi-model ensemble (MME) technique was then applied to project the SWE changes over Eurasian continent in the 21st century under the conditions of two different representative concentration pathways (RCP4.5 and RCP8.5) using eight good CMIP models out of total 26 models. The results show that the models were able to reproduce the spatial pattern of winter mean SWE in the Eurasia, i.e. the 25-year average of SWE increased from south to north and SWE in the Tibetan Plateau was much higher than those in other regions of the same latitude. However, some errors still existed in the models. For example, almost all models underestimated the maximum SWE in central Siberia, and SWE in northeastern China was also underestimated. It was found that SWE to the west of Ural Mountains and over northern part of China and Mongolia was overestimated when compared with observation. Meanwhile, only a subset of the models could produce the maximum SWE on the eastern Tibetan Plateau, and the spurious maximum SWE could be found on the western Tibetan Plateau in most CMIP5 models. The spatial and temporal characteristics of winter SWE from CMIP5 model simulations and observations were further analyzed using the Empirical Orthogonal Function (EOF) analysis, and the results suggested that only a small number of CMIP5 models could reproduce main features of the first eigenvector that reflects the decadal variation of SWE over the whole Eurasia. The second mode reflects the annual variation of SWE over the Eurasia, and only a few models (e.g., INMCM4) could reproduce the spatial and temporal characteristics of the second mode to some extent. With respect to the reference period 1981-2005, projection of SWE by the MME under the RCP4.5 shows that SWE in the northeastern Eurasia continent would increase significantly with an increase of 4.1 mm for the 25-year averaged winter SWE in the early stage of the 21st century, followed by 5.4-mm and 6.8-mm increases in the middle and late 21st century, respectively. In contrast, there would exist a decrease of SWE in continental Europe to the west of 90°E and over the Tibetan Plateau and the decrease would become more severe with time. In terms of percentage change of SWE, the region with large magnitudes was found in the northeastern Eurasian continent, where the increase of SWE could be around 5%-10%. However, no maximum centers were found in the Tibetan Plateau, Scandinavian Peninsula and East European Plain possibly because of the large values of winter SWE in these regions. Projection of SWE changes by the MME under the high emission scenario RCP8.5 shows a similar pattern with results under the emission scenario RCP4.5, but with larger amplitudes of changes in snow water equivalence.