In the context of full automation of manufacturing processes, it is important to develop predictive maintenance methodologies in every industrial field. This paper presents experimental tests carried out on some components of high-performance automatic machines. The aim is to assess and compare the effectiveness and reliability of different condition monitoring methodologies based on vibration analysis. A methodology based on recovery of the acceleration pattern of the final actuator of the mechanism under consideration seems to be particularly promising. It allows the working conditions to be judged, directly comparing the recovered output member acceleration with the functional requirements which, for this kind of mechanism, are mainly relative to the motion law. A lumped parameter model of the same mechanism is presented and its application for diagnostic purposes is shown.