Vegetation coverage is an important parameter in terrestrial ecological process, meteorological, and climatic models. By eliminating the errors from the precision of image classification and the noises of remote sensing images, and by using the actual data from fieldwork, this paper determined the maximum and minimum values of normalized difference vegetation index (NDVI), improved the sub-pixel model, and verified this model by calculating the vegetation coverage of Beijing. The results showed that the estimation value of the improved model was very close to the measurements, especially for the herbaceous plants whose vegetation types were the same but the densities were different. However, the estimation error of arborous vegetation coverage was relatively large, probably due to the effects of remote sensing image resolution, vegetation fragmentation, and mixed pixel model.