Ball mills play an important role in both energy consumption and metal wear in mineral processing plants. To maintain high operating efficiency, the material transportation inside the tumbling body has to be monitored in time. It is known that the vibration signal pattern varies corresponding to the operating state of the mill. Besides the basic vibration signature from the rotary drum and machine assembly, the tumbling of steel balls and the material are the major vibration sources. Since the steel balls and the material are unevenly distributed along the rotating axis the vibration sources are spread widely. The location of the vibration sensor has to be optimised to obtain representative signals for the process. Nine locations on the trunnion bearings and the bearing for the pinion axis have been investigated to select the best place for situating a vibration sensor. The vibration signal was picked up by an accelerometer in the form of time-domain waveform, which was firstly recorded by a DAT deck and then digitised by an oscilloscope. The digital signal processing and system identification were performed using software specially developed for an IBM compatible personal computer. The power spectra from different locations were studied and one best sensor location was recommended for picking up a representative signal from the ball mill. More sensors on different bearings are required for mapping the whole picture of the milling state.