Interturn short circuit fault is one of the most common incipient faults in transformers. To improve the diagnosis accuracy of the mild faults， a multi-feature fusion diagnosis method for transformer winding short-circuit faults is proposed considering winding short-circuit current， winding temperature， current density， and magnetic flux characteristics. A COMSOL space model of a transformer with winding short faults is established. And the comparison of simulation results with the theoretical calculations is performed， which proves the correctness of the represented model. Then， the fault simulation and characteristics of transformers with mild faults are analyzed， and the characteristic parameters from current， temperature characteristics， current density， and magnetic deviation are extracted as fault features. BP network is used to identify the fault types. The effectiveness of this method is verified by an engineering diagnosis example of a transformer.