Forest is one of the vital renewable resources for sustainable development of renewable resources; it plays an important role in global climate change, water and soil conservation, and carbon cycle in terrestrial ecosystem. Forest biomass, therefore, is now attracting attention worldwide from both scholars and policy makers. Using Landsat8 OLI images and 296 survey samples in Fujian Province, we found that the leaf biomass is negatively correlated with the reflectance of near infrared wave band and the slope of near infrared and red band. Therefore, the slope of the near infrared and red band reflectance can be used as an effective indicator for describing the differences in leaf biomass of forest. Currently, empirical models are mainly used to estimate forest biomass such as the vegetation-index models based on multiple spectra remote sensing imageries and backscattering-coefficient inversion models based on active microwave remote sensing imageries. However, most of these empirical models lack physical mechanism. Here, we established a spectral-slope based model building on the spectral characteristic of multiple spectra remote sensing of forest. We firstly used the pixel unmixing model to select the pure vegetation pixels from the images and calculated the slope (Slope_{(red, infrared)}) of the reflectivity of red band and near infrared band from image spectral curve characteristics based on the pure vegetation pixels of forest communities. We then set up the leaf biomass (LB) reversion models based on the relationship between the spectral slope and the leaf biomass by using the linear regression method to estimate the leaf biomass of coniferous forest, broad-leaved forest, and mixed forest in Fujian Province. We finally verified the results by using the in situ biomass data. The spectral-slope-based estimation algorithms for retrieving the leaf biomass of coniferous forest, broad-leaved forest, and mixed forest are LB_conifer=59.358-38.948×Slope_{(red, infrared)} (R²=70.55%), LB_broad=28.622-12.527 Slope_{(red, infrared)} (R²=68.89%), and LB_mixed=23.281-10.952 Slope_{(red, infrared)} (R²=51.75%), respectively, which indicated that our method is feasible and effective. The relationships between the leaf and above-ground biomass (AB) of coniferous forest, broad-leaved forest, and mixed forest are AB_conifer=12.079×LB_conifer-17.61 (R²=77.10%), AB_broad=23.634×LB_borad-34.124 (R²=88.86%), and AB_mixed=14.582×LB_mixed-10.789 (R²=86.78%), respectively. Our results showed that our method is suitable for the leaf biomass estimation with high correlation coefficient in Fujian Province. We also used 296 sample survey data to model the leaf biomass and the above-ground biomass including vegetation stem, branch, and leaf. The root mean square errors (RMSEs) of leaf biomass estimation model for coniferous forest, broad-leaved forest, and mixed forest are 29.2467 t/hm² (R²=66.64%), 14.0258 t/hm² (R²=61.13%), and 10.1788 t/hm² (R²=55.43%), respectively. The RMSEs of above-ground biomass estimation model for coniferous forest, broad-leaved forest, and mixed forest are 49.8315 t/hm² (R²=54.65%), 45.1820 t/hm² (R²=49.01%), and 41.5131 t/hm² (R²=38.79%), respectively.