Feed rate is an important index of combine harvester, which directly affects operation efficiency and failure rate. In order to obtain accurate feed rate information constantly in the process of combine harvester operation, a feed rate monitoring system based on header power was designed and developed, and a feed rate prediction model was established. The monitoring system was mainly composed of sensing module, vehicle terminal and mobile terminal. Specifically, the sensing module included torque sensor, hall sensor and GPS receiver, etc. The vehicle terminal displayed the collected information locally and uploaded it to the server, and the mobile terminal realized the remote monitoring of the combine harvester’s operation parameters. Based on the establishment of the linear regression model for the feed rate prediction, the study applied double threshold filtering and low-pass filtering to the torque signal. The field test result showed that the system ran stably and communicated well. It also showed that the predictive determinant coefficient of the linear regression model was 0.755, meanwhile, the filtering method can effectively filter the noise and increase the predictive determinant coefficient to 0.852, which can meet the actual need of the combine harvester’s feed rate monitoring to a certain degree.