Visible and near infrared reflectance (Vis/NIR) spectroscopy technique was applied to detect the disease severity of soybean pods anthracnose. Principal component analysis (PCA) combined with back propagation neural network (BPNN) and successive projections algorithm (SPA) combined with BPNN were used as two methods to analyze and prediction of the disease severity of soybean pods anthracnose. Data compression of SPA and learning ability of BPNN was used to achieve the detection of anthracnose severity on soybean pods. The accurate rate of identification was used to evaluate the model. The results of experiment showed that SPA-BPNN was the better calibration model and the accurate rate of detection was 90%. According to the results, SPA was a powerful way for the selection of effective wavelengths, and BPNN model could obtain the accurate detection. 