Power transformer insulation fault location is the key to improve the stability of power transformer. A Bayesian network based on power transformer insulation fault on-line monitoring method is proposed. The Bayesian network characteristic decomposition model is used to detect the insulation fault of power transformer, the high-resolution spectrum characteristic quantity of insulation fault of power transformer is extracted, the load balance analysis is carried out according to the output voltage and load difference of power transformer, the Bayesian network detection model of insulation fault of power transformer is constructed. Combined with PCI integrated information processor and relay transmission node network topology model, the on-line monitoring system design of power transformer insulation failure is realized. The simulation results show that the fault location of power transformer insulation is accurate and the visual resolution of fault is strong.