The radiation noise of the on-load tap changer (OLTC) of a power transformer changes with the mechanical fingerprint of the device during the switching operation. By placing acoustic sensors near a specific OLTC, characteristic signal acquisition and data analysis were conducted on the device through 200,000 switching experiments. Through time series analysis of the acoustic characteristics, it has been verified that four types of acoustic features (spectral centroid, energy ratio of specific frequency bands below 2000 Hz, spectral entropy, and spectral kurtosis) exhibit significant indications when major operational condition changes occur in the OLTC. These features can be used for the operation monitoring and predictive maintenance of OLTCs.