Bolt connections are widely used in aero-engine rotors and gas turbines because of their load-transferring and detachable characteristics. However, bolts are prone to looseness due to the influence of fatigue, shock, and thermal loads, which decreases the reliability of the bolted connection structure. Therefore, detecting the assembly tightness of bolted connections is critical to ensure structural integrity during the assembly phase. A high-order cumulant-gray-level image feature (HOC-GLI) method is proposed to detect the assembly tightness of bolt connections. The core of this new method is to obtain high-order cumulant images of vibration signals to eliminate noise and reserve rich nonlinear information. The amplitude distribution of the third-order cumulant reflects the energy distribution of the vibration signal in 3D images. Then, the third-order cumulant 3D images are converted to 2D gray-level images to extract the texture feature. Finally, the root entropy index of the normalized gray-level cooccurrence matrix (GLCM) based on gray-level images is used to indicate the complexity of image information. Experimental studies on six bolt connection states of the aero-engine rotor are conducted. The relationship between the root entropy index and bolted connection status is obtained to verify the effectiveness of the proposed method for assembly tightness detection in different bolt connection statuses. The result shows that the root entropy index has a decreasing trend with the loosening of the bolts, which can quantitatively detect the assembly tightness of the bolt connection structure.