Purpose -The purpose of this paper is to solve an inverse problem of structure recognition arising in eddy current testing (ECT) -type NDT. For this purpose, the space mapping (SM) technique with an extraction based on the Gauss-Newton algorithm with Tikhonov regularization is applied. Design/methodology/approach -The aim is to have a computationally fast recognition procedure of defects since the monitoring results in a large amount of data points that need to be analyzed by 3D eddy current model. According to the SM optimization, the finite element method (FEM) is used as a fine model, while the model based on an integral method such as the volume integral method (VIM) serves as a coarse model. This approach, being an example of a two-level optimization method, allows shifting the optimization load from a time consuming and accurate model to the less precise but faster coarse surrogate. Findings -The application of this method enables shortening of the evaluation time that is required to provide the proper parameter estimation of surface defects. Research limitations/implications -In this work only the specific kinds of surface defects were considered. Therefore, the reconstruction of arbitrary shapes of defects when using real measurement data from ECT system can be treated in further research. Originality/value -The paper investigated the eddy current inverse problem. According to aggressive space mapping method, a suitable coarse model is needed. In this case, for the purpose of 3D defect reconstruction, the reduced VIM approach was applied. From a practical view point, the authors demonstrated that the two-level inversion procedures allow saving of up to 50 percent CPU time in comparison with the optimization by means of regularized Gauss-Newton algorithm in the same FE model.