Abstract: Abstract: Leaf area index (LAI) is not only an important parameter of biomass estimation, but also one of the most important structural parameters for the quantitative analysis of the land ecological system's energy exchange. This paper was designed to find a method to estimate LAI, which was accurate, rapid, large scale, and not damaging. In the remote sensing estimation of leaf area index (LAI), the most commonly used methods were based on the statistics. However, it has significant limitations and had difficulty dealing with the problem of "the same thing with different spectrum, and the same spectrum but different thing" for those models. Based on the physical structure of the ground component, this study developed the linear mixture model for forest LAI estimation. It can not only deal with the difficulty of spectral discrimination, but also was simple, feasible, and general. The minimum noise fraction (MNF) method, which can eliminate the correlation between the bands of TM images and increase the quality of endmembers, was employed to convert the TM image into its principal components. After that, endmenbers were obtained from the image itself and the endmembers were regarded as the extremes in the triangles of an image scattergram. An unconstrained least-squares solution was used to un-mix the spectral image into fractions, and the vegetation cover percent was obtained from it. Then, according to the relationship between vegetation cover percent and the LAI, we were able to extract LAI from the remote sensing imagery successfully. Moreover, the canopy model of multiple scattering was applied to estimate the accurate LAI. Finally, four endmembers (green vegetation, soil, water, and non-photosynthetic vegetation) were selected, and an unconstrained least-squares solution was used to un-mix the spectral image into fractions. The average error was 0.0028, and the quality of fraction images was better. The results shows that the method that combined the linear mixture model with the canopy model could estimate the forest LAI accurately. In the study area, there was a strong correlation between the observed value and the predicted value. The coincidence degree of the model was 82.19%, and the RMSE was 0.368.