Early detection of early blight on tomato leaves using NIR hyperspectral imaging technique based on modified gram schmidt (MGS) model and Bayesian logistic regression (BlogReg) were studied. Hyperspectral images of 70 infected and 80 healthy tomato leaves were acquired by hyperspectral imaging system in the spectral wavelength of 874~1734nm. Spectral reflectance of 30×30 pixels from region of interest (ROI) of hyperspectral image was extracted. Least squares-support vector machine (LS-SVM) model based on the full wavelength was established to detect early blight. Five (911nm, 1409nm, 1511nm, 1609nm, 1656nm) and nine wavelengths (901nm, 905nm, 908nm, 915nm, 918nm, 1123nm, 1305nm, 1460nm, 1680nm) were selected by MGS and BlogReg, respectively. Then, LS-SVM and linear discriminant analysis (LDA) models were built based on these effective wavelengths. Among these models, the correct classification rates were 93%~98% in calibration set and 96%~100% in prediction set, respectively. The result indicated that it was feasible to detect early blight on tomato leaves by using NIR hyperspectral imaging technique.