Forced by NCEP reanalysis based land surface climate data, NCAR CLM3 was run in an offline way with 6 parameterization schemes (CLM3 default, Douville1995, Roesch2001, Wu2004, Yang1997, Niu2007) for calculation of snow cover fraction (SCF) to investigate the performance of these SCF parameterization schemes in simulating the seasonal variation of snow cover. In comparison with the observations derived from NOAA AVHRR, under the framework of NCAR CLM3, SCF schemes like CLM3 default, Douville1995, and Roesch 2001 underestimate SCF over vast areas, snow lines simulated in these three schemes are located to the south of that in the AVHRR observation, especially in autumn when snow pack begins to establish; Wu2004 scheme also underestimates SCF in the Eurasian continent in autumn; Yang1997 scheme slightly overestimates SCF, especially along the southern border of the snow covered area; Niu2007 scheme which accounts for the seasonal variation of snow density overcomes the positive bias of SCF simulated by Yang1997 scheme to some extent. At the end of spring, snow lines simulated by all the aforementioned 6 schemes are located to the north of the AVHRR observation. Over most land areas with relatively small variation of topography, Niu2007 scheme performs the best. The frequency of low SCF (less than 0.2) or high SCF (more than 0.8) are dominant in both the AVHRR observation and simulations of all the 6 parameterization schemes, moderate SCF (between 0.2 and 0.8) are rare phenomena.