Abstract: Nondestructive evaluation of ferromagnetic material is most commonly performed by magnetic flux leakage(MFL) techniques, and the key element is to describe the characters of defects from MFL inspection signals. A novel method for the reconstitution of 2-D profiles is presented based on least squares support vector machines (LS-SVM) technique, and particle swarm optimization(PSO) is adopted to optimize the model parameter of LS-SVM. The input data sets of SVM is MFL signals and output data sets is 2-D profiles parameter, the mapping relationship from MFL signals to 2-D profiles of defects is established. The least squares method is introduced into network learning, the training data sets are composed of experiment data sets and simulation data sets, the testing data sets are artificial crack defects. The reconstitution of 2-D profiles of artificial crack defects in the magnetic flux leakage testing was implemented by this algorithm. Comparing with the reconstitution results of BP network and GA-LS-SVM, the results show that LS-SVM possesses quick speed、high accuracy and very good generalization ability , and it is a good way for the quantization of the MFL testing.