Abstract: For the existent mixed pixels which appear in using MODIS for area estimation of the small water body，2-D scattered points-classification NDVI method is proposed.This method firstly analyzes band 1 and band 2 of MODIS data through 2-dimensional visualization，secondly selects sample point sets of ground features distinguished by vision to train BP Neural Network，and then discriminates and classifies pixels in the test areas to get mixed pixels，finally estimates water body area by using linear spectral mixed pixels decomposition model.Above-mentioned method is applied to estimate area of the Qionghai Lake in Sichuan province and discuss the error in detail.The results show that 2-D scattered points-classification NDVI method，has a better adaptability than some other water information extraction method and a higher precision in quantitative calculation of small water area，is applicable to real-time monitoring of small size water such as small lakes and reservoirs.