ABSTRACTEurasian snow cover fraction (SCF) prediction is analyzed using the recently developed National Centers for Environmental Prediction (NCEP) Climate Forecast System version 2 (CFSv2) monthly retrospective forecasts for 1983–2009. The CFSv2 is generally capable of reproducing the observed Eurasian SCF seasonal cycle and climatology. This study focuses on the prediction skill and predictability of Eurasian SCF in snowmelt and snowfall seasons because the intensive variability occurs in the two seasons. The CFSv2 reasonably predicts the interannual variations, long‐term trend and leading pattern in snowmelt season several months ahead. In comparison with the snowmelt season, the CFSv2 shows a better prediction skill in climatological values but a worse skill in the interannual variability in snowfall season. In addition, the forecasted downward trend of SCF in the snowfall season is opposite to that in the observation. The biases of Eurasian SCF in the snowmelt and snowfall seasons are significantly related with those of temperature and precipitation in the CFSv2. The forecasted cooler and wetter atmosphere is suggestive of the overestimation of the mean SCF. Meanwhile, the underestimation in the variability of both temperature and precipitation in the CFSv2 may be the important factor for the underestimated variability of SCF, especially for the damped variability of SCF in the snowfall season. Generally, the CFSv2 shows a higher and more stable prediction skill after late‐1990s than before in the two seasons. The change in the initial condition in the CFSv2 and the observed SCF in late‐1990s might be the plausible reason for it.