Snow is a unique environmental factor in the seasonal snow-covered area. The snow parameter inversion using remote sensing data has great significance on the researches of regional soil moisture forecast, hydrology, climate, etc. The problem of great error exists in the binary classification method commonly used in snow retrieval research. According to the unique geographical environment and the surface type in the black soil region of Songnen Plain, the MODIS images were chosen as data source, and the OLI images were regarded as “true value”data. Then the linear regression relation model between snow cover fraction from MODIS and NDSI was established in black soil region of Songnen Plain. The results showed that the adaptability of the MOD10A1FSC data was weak in the study area. The FSC data was 80.21% which had a big difference compared with the snow cover fraction of OLI images (87.71%) at the same time phase. The correlation coefficient between the FSC data and OLI images was only 0.58. The snow cover fraction of the inversion model built in the study was 85.28% which was close to that of OLI images at the same time phase. The correlation coefficient between the snow cover fraction of the inversion model and OLI images was 0.66. In addition, compared with the MOD10A1FSC data, the error statistics results, including root mean square error and mean absolute error of the inversion model were decreased significantly. The estimation model of snow cover fraction based on sub-pixel improved the monitoring accuracy of snow cover in black soil region of Songnen Plain to a certain extent, which can better satisfy the reality requirement to the current snow retrieval research. The research result provided a scientific basis for the soil moisture forecast in spring and agricultural cultivation in this region.