Abstract: To explore a properly method for characterizing stress concentration degree quantitatively by metal magnetic memory testing (MMMT), tension-compression fatigue tests of specimens with different stress concentration factors made of 42CrMo steel were carried out. Both normal and tangential component of magnetic memory signals of specimens under different fatigue cycles were measured by magnetic memory apparatus. A back propagation neural network (BP neural network) was built to distinguish the stress concentration degree, whose input eigenvector was the feature extracted from magnetic memory signals. The results showed that the BP neural network could be used to recognize stress concentration degree of specimens quantitatively.