In order to improve the calculation accuracy of theoretical line loss for distribution system, a novel calculation method based on hybrid learning algorithm(HLA), which is composed by generalized regression neural network(GRNN) and particle swarm optimization(PSO), is proposed. The theoretical line loss calculation is abstracted into the multi regression problem (MRP), of which the input and output vector are selected as various impact factors and theoretical line loss values. The sample set is constructed for HLA training. In GRNN training process, the method can overcome the difficulty for selecting optimal training parameters by PSO. The experiments show that, the HLA method has better performance compared with the traditional one.