Abstract: The study proposed how to extract the vegetation coverage precisely on a large scale and in short time using the MODIS Normal Different Vegetation Index（NDVI) and the multi-spectral image Landsat TM without ground measurements：Get the vegetation coverage making use of the two method of empirical model and dimidiate pixel model.Then the results were compared among different models on basis of the TM coverage with higher precision.The conclusion is that in the study area where vegetation coverage is distributing equably，regression models are better than dimidiate pixel model in extracting the vegetation coverage in larger sample scale，especially to the nonlinear model，besides，the errors caused by noise and geometric registration can be reduced effectively.

Key words: NDVI, Vegetation coverage, Sample scale, Dimidiate pixel model