A version of the Penn-State/NCAR mesoscale model (MM5) has been running a continuous Four Dimensional Data Assimilation (FDDA) in realtime, via the nudging technique, over the midwest United States for the past several months. A snow prediction algorithm was implemented in the realtime system this winter with the aim of reducing the model bias over a week to 10-day integration and providing more realistic land-surface characteristics for snow-covered regions, thereby improving the surface-energy balance and representation of local mesoscale circulations. The snow prediction scheme is consistent with the existing model's physics, and in particular with the microphysics and the surface energy budget modules. The microphysics predict rain water and cloud water species and undergo phase change at 0C. Preliminary results show improved statistics when compared to a parallel run without snow prediction. The verification of the surface (2m) temperature and wind (10m) fields show improved mesoscale circulations such as slope flows in complex topography. The implementation of a snow prediction in a realtime framework has provided us with challenges ranging from meso-beta scale initialization on 3.3-km resolution to having a proper snow data to initialize the coarser (30-km) resolution at cold start. The practical experiences and statistical evaluations of the system's performance with and without snow prediction during this winter will be contrasted.