Abstract: The experiment acquired Vis-NIR of navel orange surface, which were sprayed with different concentration of omethoate pesticide. The prediction models of partial least squares (PLS) pesticide contamination were established using the four different spectral pretreatment methods of multiplicative scatter correction (MSC), standard orthogonal variable transformation (SNV), first derivative (FD) and second derivative (SD) by selecting the spectral range 430-1000 nm, 1 000-1 800 nm and 430-1 800 nm respectively. Comparing the results of experiment, it shows that the optimal model is obtained in the range of 430-1 000 nm and by the spectral data preprocessing method of FD. In the validation set, the correlation coefficient Rpred between the actual category and the predicted category of omethoate contamination degree on navel orange surface is 0.9817 and the root mean squared error of prediction samples (RMSEP) is 0.1564.