As a core device in power systems, transformers have winding faults accounting for over 80% of all transformer faults, which seriously threaten the safety of power grids and social-economic production. This paper proposes a new method for transformer winding deformation diagnosis based on a time-domain algorithm: by collecting voltage and current recording data during system short-circuit faults, the differential method is used to effectively attenuate the interference of aperiodic components on calculations; then, compensation and correction are performed for the discretization error introduced by “replacing differentiation with difference”. Finally, the transformer short-circuit impedance is accurately solved, and the winding deformation state is determined based on the impedance change. Verified by MATLAB simulation (taking the 50MVA main transformer of Liupanshui Substation in Guizhou as the object) and on-site fault recording data, this method can stably output the short-circuit impedance result within 1 power frequency cycle (0.02 s). The simulation error is only 0.018%, and the calculation error of on-site data is 2.35%, which verifies the accuracy, rapidity, and engineering practicability of the method.