Snow surface temperature monitoring has a very important significance for climate change research, snowmelt runoff forecasting and snowmelt flood prevention. The purpose of this study is to use MODIS data to retrieve the sub-pixel snow surface temperature. In this paper,firstly the linear coefficient of the Plank function in low temperature environment was simulated,then the sub-pixel snow surface temperature retrieving formula was deduced,and a surface average emissivity estimation method was proposed based on the pixel snow cover fraction and forest coverage area data. The results indicate that compared to the traditional component temperature retrieving method,this algorithm is simple and easy to operate and it can improve the MODIS snow surface temperature retrieval accuracy in a certain extent,especially for the mixed pixel area.