Aiming to the problem of automatic modulation classification of the existing digital signal，a classification method based on Kernel Logistic Regression(KLR) is developed.This method is primarily used in economic，medical science and speech process etc，while seldom applied in the field of communication signals. The characteristic parameter of high-order cumulants of the signal is used for training data and testing data.The classification is performed adopting the frequently-used decision tree method. The proposed method is compared to the modulation classification method based on Support Vector Machine(SVM) through simulation experiments. The results indicate that the proposed method is qualified to do the work.Under low SNR(0 dB)，the performance of classification is higher than that based on SVM; while under 5dB，the correct recognition rate is above 90% based on KLR.