The gearbox fault diagnosis is the effective measure to avoid the sudden accident and ensure the normal, stable operation of large-scale wind turbine. Aimed at the real-time vibration monitoring of large scale wind turbine gearbox, its vibration condition was analyzed and fault diagnosed. The collected signals were researched by the time-domain method, and the signal time domain attribute was achieved. Each information attribute was reduced by using knowledge of neighborhood rough set, and then combined with support vector machine (SVM) for sample training. Moreover fault diagnosis of the test data is realized.