Based on the canopy hyperspectra and derivative spectra of typical hardwoods in Mudan valley, Changbai Mountain, red edge characteristics were used to determine leaf coloring date and to establish regression models of spectra and leaf coloring date. The results show that the canopy hyperspectral reflectance can accurately reflect the change of vegetation autumn leaf coloring date. The states of the leaves of typical hardwoods can be divided into three categories: the dry leaves not falling after leaf coloring date, the leaves falling out before leave full coloring date, and the leaves falling after leaf full coloring date. The first derivative spectral curves and hyperspectral curves take clear advantage over the specific date of leaf first coloring date and leaf full coloring date. The variances of inversion models of red edge parameters and leaf coloring date are all above 0.9, and each vegetation is suitable for different forms of fitting equation. The research shows important theoretical significance and extensive application prospect on using quantitative remote sensing to monitor mountain autumn phenology.