This study investigates the use of acoustic emission (AE) for in situ monitoring of surfaces sliding under starved conditions until failure due to scuffing mechanism. Reciprocal lubricated sliding tests having flat-on-flat set-up have been carried out for a cast iron–steel tribo-pair at a constant load of 600 N and a frequency of 6 Hz. According to the friction behavior, three regimes of events have been specified; steady-state, pre-scuffing, and scuffing. Acoustic signals for these regimes have been decomposed with wavelet packets, and sub-band energies have been chosen as features. The classification is performed using support vector machine. Experimental results reveal the feasibility of automatic detection of surface pre- and scuffing states using acoustic emission.