The invention relates to the technical field of transformer fault on-line monitoring, in particular to a transformer maintenance decision method based on a hybrid model. According to the method, a hybrid modeling mode of an RBF neural network and a PHM proportion failure model is adopted, so the advantage that the proportion failure model can be maintained according to needs is fully developed compared with a traditional maintenance mode, the screening function of the RBF neural network method and an MIV method on characteristic quantities is fully utilized, and accordingly the accuracy and reliability of maintenance decision schemes of the proportion failure model(PHM) are further improved.