Abstract: Near-infrared diffuse reflectance spectroscopy ( NIDRS) combined with principal component analysis ( PCA) and least squares support vector machine ( LSSVM) were used for the identification of transgenic rice. PCA was used to analyze the spectral spatial distribution of rice. Different spectral preprocessing methods including 5-point smoothing, multivariate scatter correction ( MSC) and standard normal variate transformation ( SNV) combined with LSSVM were used to build and optimize qualitative models.The grid search algorithm was employed to obtain the optimal solution of the penalty factor ( c) and the parameters gamma ( g) of RBF kernel. The correct recognition rate ( CRR) were used to evaluate models. The results showed that MSC combined with LSSVM could be used to establish the qualitative identification model of transgenic rice. The CRR of the optimal model was 97.50%.The method was expected to be an auxiliary method for rapid detection of genetically modified foods.