Aiming at the problem that the single component of magnetic gradient tensor has insufficient ability to recognize oblique magnetized multi-targets and is greatly affected by the carrier attitude, a magnetic target recognition method based on singular value decomposition (SVD) of magnetic gradient tensor is proposed. Through the singular value decomposition of the magnetic gradient tensor matrix, the maximum of the singular value is extracted as the magnetic target recognition feature, which enhances the recognition ability of inclined-magnetized multi-target, proves the tensor invariant property of magnetic gradient tensor singular value, and overcomes the influence of carrier attitude change on magnetic gradient tensor recognition. The simulation and experimental results show that the method can effectively distinguish and identify multiple targets with oblique magnetization.