A new method was studied to detect pesticide residues in lettuce leaves rapidly, accurately and nondestructively. In this paper, four groups of lettuce were used as experimental samples, the first group was sprayed with acetone, the second group contained dimethoate (volume ratio between omethoate and acetone is 1∶1000), the third group contained dimethoate (volume ratio between omethoate and acetone is 1∶500), the last group of lettuce leaves dimethoate (volume ratio between omethoate and acetone is 1∶100). Totally 384 samples of four varieties were scanned by using the NIR hyperspectral imaging system (871.61~1766.32nm). The region of interest (ROI) in hyperspectral image of samples was selected, and the mean spectra of all pixels in the region of interest was calculated. At the same time, optimal image selection was carried out by principal component analysis (PCA). The first principal component (PC1) image and the second principal component (PC2) image were used for texture features analysis. Among the processing of spectral data, successive projections algorithm (SPA) and principal component analysis (PCA) were used to obtain characteristic wavelengths. Finally, Knearest neighbors (KNN), Mahalanobis distance(MD), Fisher linear discriminate analysis (FLDA) algorithm were used for model establishments respectively based on spectral feature and the combined features in full and characteristic wavelength. In all models, the performance of FLDA based on the combination of texture and spectral features by SPA has its superiority in classification recognition with the training rate of 98.90% and prediction rate of 100%. The results show that it is feasible that NIR hyperspectral image with data fusion is used to discriminate the grade of pesticide residue.