Underwater robot (RemoteOperatedVehicle, hereinafter referred to as the ROV) underwater hovering is a very important shipment.In order to achieve the underwater hovering control, we design a hovering control method based on multi-sensor information fusion. The ROV respectively calculated by accelerometer and gyroscope and depth for the body of the shaft, attitude Angle and depth of acceleration signal of the gyroscope data by quaternions, get the body attitude Angle in real time, through the weighted sliding filter of gyroscope and accelerometer data beforehand. After using kalman filter for accelerometer and gyroscope data fusion, get real-time acceleration. Finally, BP neural network to forecast the depth and the attitude Angle, and the prediction of depth and the attitude Angle into the PID controller as the control strategy. The fusion method can effectively complete the hovering control, and the experimental results show that this method can achieve higher stability and real-time performance, has a broad prospect in the hovering control.