This paper describes time-frequency analysis of the friction coefficient with discrete wavelet transform. Ball-on-disk experiments were conducted with a steel ball, an aluminum disk and ethanol as a lubricant. The coefficient of friction was rather stable and low during initial sliding before it suddenly increased by seizure caused by loss of the lubricant. While no indication of transient change could be found before the seizure occurred in the original friction and ball displacement signals, the wavelet components of the friction coefficient clearly showed the local disturbances during several cycles before the sudden increase. The results suggest that the wavelet transform is potentially applicable to the prediction of catastrophic changes in friction.