95.3%。 A method for maize cultivars identification was proposed, by using the combination algorithm of K-L transform and least square support vector machines (LS-SVM)．Firstly, labeling algorithm and multi-scale wavelet analysis were used to acquire single corn kernel images; secondly, K-L transform was employed to extract the features of corn kernel images; at last, the method of LS-SVM with a binary tree architecture was developed to identify maize cultivars by using feature data．According to the result of test, the reasonable state space dimension (L=3) is acquired, and this method gives a correct recognition rate of 95.3%．