Gearbox is one of the most important components widely used in rotary machines and its health status is the key factor for the stable operation of the machinery. Hence, the condition monitoring and fault diagnosis of gearbox is of great significance. A new gearbox fault diagnosis method based on support vector regression is proposed. Firstly, the features that can reflect the health status of the gearbox are extracted, and then the problem of classification is transferred to regression problem. Finally, a new support vector regression decision mechanism is constructed and applied to the diagnosis of gearbox. It effectively avoids the problem of equal votes in voting decision organization. Comparing to Artifical Neural Network (ANN), the proposed method converges fast and has better generalization ability.