Metal magnetic memory (MMM) technology is a nondestructive testing method, which can conduct effective diagnosis of early microscopic damage of ferromagnetic material. In order to eliminate the uncertain influence factors of magnetic memory signal and improve the accuracy of damage state recognition, the magnetic gradient tensor and magnetic field signal vertical distribution feature analysis methods are introduced. Firstly, the magnetic gradient tensors of the MMM signals on the crack fracture zone and stress concentration zone are measured using a triaxis magnetometer. From the measured results, both the plane and vertical characteristics of the MMM signal distributions are obtained. To remove the influence of the measuring direction selection on experiment results, a new magnetic field invariant characteristic parameter  the magnetic total gradient modulus is introduced to determine the location and boundary of the damage and damage zone. Then, the vertical distribution features of the magnetic total gradient modulus are acquired by measuring the plane distribution of the magnetic total gradient modulus under different lift offs. Finally, the difference of the vertical distribution features of the magnetic total gradient modulus at the boundaries of different types of defects is analyzed. Theoretical analysis and experiment result show that as the lift off increases gradually, the attenuation velocity and amplitude of the magnetic total gradient modulus on the boundary of the crack are far greater than the ones caused by stress concentration, and the vertical distribution features of the magnetic total gradient modulus can be used to identify the defect state effectively.