Misalignment is a more commonly encountered malfunction. It generates reaction forces and moments in the coupling and causes the rotor vibrates excessively. Generally, the vibration responses of the shaft misaligned rotor are comprised of nonlinear components. Some widely focused linear feature extraction methods are incapable to extract the key nonlinear information. Nonlinear Output Frequency Response Functions (NOFRFs) weighted contribution rate (WNOFRFs) is a nonlinear feature extraction method developed from NOFRFs, which is a convenient frequency-domain analytical method for nonlinear systems. WNOFRFs have been deeply researched and applied in rotor fault diagnosis. The present research introduces the concept of Kull Back-Leibler (KL) divergence to WNOFRFs to derive an improved nonlinear feature extraction method, the improved WNOFRFs based on KL divergence (INWKL). As a novel indicator extracted by NWKL, KR is denoted and employed to identify rotor shaft misalignment. Dynamic simulation and experimental verification indicate the INWKL method is effective and superior. And KR is more favorable due to its higher sensitivity and comprehensive representation of nonlinear characteristics of rotor shaft misalignment. Moreover, it has strong stability to speed fluctuation and can detect misaligned angles of the rotor system quantitatively.