LAI retrieval from large scale area quickly and accurately is an important research in remote sensing fields. In this paper，a model is presented to estimate LAI of reed canopy from Landsat-5 TM image data. Firstly，the model classifies the background of reed canopy into soil and water，then calculates and output a lookup table (LUT) by use of FCR model. Following it，LAI mapping was conducted based on the BP neural network model，which was trained using the data of actual measurement and LUT. The results indicate that the method has strong nonlinear fitting ability，and is able to increase the accuracy of LAI results through reducing the background influence from background spectrum.