Walnut is an important dry fruit and woody oil crop in China, and it has significant meaning to establish a rapid, nondestructive testing method for identification and classification of walnut kernel varieties in walnut processing industry. Near-infrared diffuse reflection spectroscopies of 200 walnut samples of four species were adopted to establish models for rapid and nondestructive classification. The spectral region of walnut samples was ranged from 3 800 cm -1 to 9 600 cm -1 . The spectra data of walnut were processed using the multiplicative scatter correction (MSC) and the standard normalized variate (SNV) methods. Principal component analysis (PCA) was used to reduce the dimensionality of spectra data. The cumulative contribution rate of the first five main components reached 99.21%, which were selected as variables for modeling. Totally 100 walnut samples were selected as training set by random sampling method. The NIR classification model of walnut kernel varieties was built based on support vector machine (SVM) method, and grid search method was used for searching the best parameter. The built model was tested by the rest 100 walnut samples of four species, and the results showed that the correct recognition rate of the model reached 96%. The analyzed results indicated that the NIR classification model could provide a feasible method for rapid and nondestructive identification of walnut kernel varieties.