This paper introduces a new tool breakage diagnosis technique by using a support vector machine (SVM) in face milling. From the viewpoint of frequency domain, the paper is focused mainly on the diagnosis with spectrum of cutting force signals. With the spectrum, the SVM is learned to adapt the diagnosis. As the substantial benefits in classification, the system, joined the spectrum input and the SVM learning, is capable of responding in real-time to diagnose automatically when a tool fracture occurs even under the varying cutting conditions, and is really admissible to monitor the machining tool with or without breakage. As for the experimental results, they show that this new approach could sense tool breakage in a wide range of face milling operations.