Late blight has become a severe disease of potatoes that can decreasethe quality of the product and yield.The primary aims of the current study were :(1)to study the feasibility of using hyperspectral data to detect late blight in potato tubers;(2)to evaluate the performance of different spectral data preprocessing methods;(3)to select effective wavelengths for disease detection by chemometric methods in spectral domain;(4)toachieve the best classification model after comparing different models.In this study,hyperspectral datafrom 375 to 1012nm were acquired.Spectral and images of samples in three grades(healthy,mild,severe)were extracted by ENVI software.The performance of twelve different spectralpreprocessing methodswere compared inpartial least square(PLS)models.The sensitive wavelengths for late blight that selected by p rincipal componentanalysis(PCA),successive projections algorithm(SPA)and X-Loading Weight(X-LW)were used to build models of linear partial least square discriminant analysis(PLS-DA),linear fisher discriminant analysis(F-DA)and nonlinear least squares-support vector machine(LS-SVM).The results showed that F-DA model in full spectral range attained the best prediction accuracy of 95.08%to classify samples into three grades: healthy,mild or severe.And SPA+F-DA model was more suitable for practical application with the accuracy of 90.16%due to its simlpeness.It also provides an automatic and feasible approach to detect late blight in potato tubers using hyperspectral method.