The invention discloses a statistical and classifying method for loads of a large regional power grid and belongs to the field of regional power grid load modeling. According to the technical scheme, the method comprises the following steps: firstly, performing first-layer clustering calculation on a transformer substation group on the basis of a bubble sorting method, thereby realizing the rough classifying of the transformer substation group, and then, performing second-layer clustering calculation on each class of transformer substation which is roughly classified via a kernal-based layer-by-layer clustering method, thereby realizing the fine classifying on the basis of the rough classifying. According to the method, on the basis of the load structure of the transformer substation, a kernal-based layer-by-layer clustering technique is utilized to naturally classify according to sample characteristics, so that the transformer substation group which cannot be linearly classified can be linearly classified in a higher space through characteristic mapping. Meanwhile, an application of a kernel function is combined for effectively preventing the curse of dimensionality of the higher space, the problem of difficulty in transformer substation classifying is accurately solved, and the effective statistics and classification for the loads of the large regional power grid are realized.