Near infrared (NIR) spectroscopy was applied for the fast determination of proline in oilseed rape leavers. The oilseed rape leaves were treated by herbicide,248 samples were collected for NIR spectral scanning within the wavelength region of 1100-2500 nm. Smoothing way of Savitzky-Golay with 7 segments,standard normal variate (SNV) and second derivative were used as preprocessing methods of spectral data before the calibration stage. Partial least squares (PLS) analysis was applied as calibration method as well as a way to extract the new eigenvectors which could be used to represent the most useful information of original spectra and compress the spectral dimensionality. The selected new eigenvectors were used as the input data matrix of back propagation neural network (BPNN) and least squares-support vector machine (LS-SVM) to develop the BPNN and LS-SVM models. The calibration set was composed of 186 samples,whereas 62 samples in the validation set. The results indicated that LS-SVM model achieved the best prediction performance,and LS-SVM model outperformed PLS and BPNN models. The correlation coefficients (r),root mean square error of prediction (RMSEP) and bias by LS-SVM model were 0.995,0.041 and 0.000,respectively. The overall results demonstrated that NIR spectroscopy combined with LS-SVM model could be successfully applied for the determination of proline in oilseed rape leaves treated by herbicide.