The fault diagnosis is important in continuously monitoring the performance and quality of manufacturing processes. Overcoming the drawbacks of threshold approach, artificial neural network may extract the symptom of the faults through learning from the samples, but it is difficult to design its structure. Moreover, it needs a large numbers of samples in practice. In this paper, support vector machine approach was proposed to overcome these limitations based on statistics learning theory, and a new fault diagnosis system is developed. The experimental results showed that it is an efficient and practical on-line intelligent monitoring system for the stamping processes.