Abstract: In this study, soybean was taken as the research object, and the hyper-spectral data of soybean canopy at the filling stage were measured by spectrograph and various hyper-spectral vegetation indexes were calculated. We used Linear Regression（LR）, Support Vector Machine（SVM）, Back Propagation Neural Network（BPNN） and Random Forest（RF） to established the mathematical model of the relationship between hyper-spectral vegetation index combination and soybean above-ground biomass. The results showed that the determination coefficients（R~2） of estimating AGB model based on LR, SVM, BPNN and RF were 0.59, 0.71, 0.73 and 0.76 respectively. The Root Mean Square Error（RMSE） was 2 559.0, 481.1, 1 194.6 and 805.2 kg·ha^-1 respectively. The relative analysis error（RPD） was 1.22, 1.55, 1.87 and 1.92 respectively. The prediction accuracy of models based on RF was more reliable than LR, SVM and BPNN models. Therefore, the above ground biomass of soybean can be estimated more accurately by using RF models.