Abstract: Leaf spectrum is very important to estimate vegetation biochemical parameters. However, the spectrum obtained from remote sensing is pixel and canopy spectrum, therefore, it is necessary to transform the spectrum from canopy level to leaf level when estimating leaf biochemical parameters by remote sensing data. The scaling conversion function during downscales from pixel spectra, canopy spectra to leaf spectra was derived according to principles of geometrical optics model in this paper. First, PROSPECT model was used to simulate leaf spectra. Then, with the other parameters unchanged, the canopy spectra was simulated under different leaf area index(LAI) and leaf spectra by 4-scale model, and the relationship between leaf reflectance and sunlit canopy reflectance was found. Finally, two lookuping tables were established based on LAI to achieve transformation from canopy spectra to leaf spectra. One is used to describe the relations between the probability of observed sunlit canopy and observed illuminating background. The other is for scattering factor calculation. The result indicates that leaf spectra can be well converted from canopy spectra using 4-scale model. The proposed method is very effective and useful.