The condition detection of wind turbine is always an important issue which attract more and more attentions because of the rapid development of wind farm. And the on-line data analysis is also difficult since a lot of measured data is collected.<p> </p> In this paper, the abnormal data rejection and normal condition evaluation of wind turbine is processed. At first, since there are large amounts of abnormal data in the normal operation of wind turbine, which is probably caused by fault, maintenance downtime, power-limited operation and failure of wind speed sensor, a novel method is proposed to reject abnormal data in order to make more accurate analysis for the wind turbine condition. The core principle of this method is to fit the wind power curves by using the scatter diagram. The data outside the area covered by wind power curves is the abnormal data. The calculation shows that the abnormal data is rejected effectively. After the rejection, the vibration signals of wind turbine bearing which is a critical component are analyzed and the relationship between the vibration characteristic value and the operating condition of wind turbine is discussed. It will provide powerful support for the accurate fault analysis of wind turbine.