The invention belongs to the motor equipment monitoring and fault diagnosis technical field and discloses a motor abnormal state online early warning method. The method includes the following steps that: data of parameters related to the state of a motor are periodically acquired from the supervisory information system (SIS) of a thermal power plant, and a prediction model for motor three-phase winding temperature is established; the distribution characteristics of motor three-phase winding temperature variance is put into statistics on the basis of the acquired data; and online early warningis performed against the abnormal state of the motor according to different stages on the basis of the motor three-phase winding temperature variance and the prediction model. With the method of the invention adopted, the timeliness and precision of online early warning operation can be significantly improved; purposeful spot inspection tasks can be effectively performed; and operation and maintenance management costs can be greatly reduced with the normal and stable operation of a unit ensured. The method is particularly suitable for application conditions such as various types of middle- andlarge-scale thermal power plants.