To detect and prevent potato late disease, the peroxidase (POD) activity of potato late-blight leaves was predicted by spectroscopic techniques, and the prediction of potato late-blight disease was realized based on POD enzyme activity. The spectral reflectivity and POD enzyme activity of potato leaf samples in different temperature, humidity and inoculation time conditions were collected and measured. And the Mean Centering method is ultimately chosen, which is used to eliminate the error of the original spectral data. In order to reduce the complexity of the model, RF, SPA and CARS algorithms were used to filter the wavelengths, and the results showed that the partial least-square regression (PLSR) prediction model was established by using the spectral data at 72 characteristic wavelengths which are extracted by the CARS algorithm was the best. The coefficient of determination Rp2 of the prediction set is 0.958 1, and the root means square error RMSE_p is 25.698 6 U·(g·min)^-1. Finally, the RBF radial basis network was used to fit the relationship between POD enzyme activity, temperature, humidity and inoculation time and established a kinetic model of POD enzyme activity. So the prediction of the disease period of potato late blight based on POD enzyme activity was further realized. The results proved the feasibility of using spectroscopy to rapidly determine POD enzyme activity to predict potato late blight.