Based on least squares—support vector machine (LS—SVM), the effective wavelength (EW) in visible/near infrared (Vis/NIR) region was proposed as a new approach for the variety discrimination of pears. 210 pear samples were used for the calibration set, while 30 samples for the validation set. After partial least squares (PLS) analysis, the EWs were selected according to the X-loading weights and regression coefficients, and an EW—LS—SVM model was developed for the variety discrimination. This model was compared with EW—BP-ANN model by using back-propagation artificial neural network (BP-ANN).Results showed that the same recognition accuracies (100% for the calibration set, 93.3% for the validation set) were obtained for EW—LS—SVM and EW—BP-ANN models, respectively. Studies show that it is feasible to use EW—LS—SVM model for the variety discrimination of pears.