The classification and synthesis for load characteristics is the effective way of settling the time-variation problem in composite load modeling. Based on the random process correlation theory, a new classification and synthesis method for dynamic load characteristics is presented. It uses the measured response space as the character vector space of load characteristics. Firstly, the measured response of each sample is standardized in time coordination; secondly, the correlation coefficients between every two samples in measured response space are computed; thirdly, the system clustering method is used to make the load characteristics classification according to the value of correlation coefficients; at last, the clustering center of each classification is gained by computing the gravity center of all the samples in this classification, and by identifying the clustering center of each load characteristics classification, the synthetic load model of each classification is gained. With a project instance, the correctness, effectiveness and convenient realization of the new method are proved.