This paper describes a real application of fault diagnosis based on statistical signal processing. The system monitories several state variables of the cutting process of a multi-tooth machine tool. However, the feed drive current has been chosen to detect and diagnose the most frequent faults. Experimental data have allowed to define statistical behaviour of the variables for non-fault conditions, tool wear and breakage. The goal of the system is to optimize the life-time of each tool, while ensuring dimensional tolerance in the product. The machine tool that has been monitored, is a complex machine with five tool-holders and more than 250 inserts. This machine tool is an important element in the production line of crankshafts for an automobile industry.