In the framework of non-destructive eddy current testing, an inversion method which is relying on a metamodel-based optimization method is proposed: a fast optimization with a particle swarm algorithm is ensured by replacing the forward computations by a metamodel. This metamodel is obtained by radial basis function interpolation over an efficient database generated by sequential design. The metamodel is then connected to a particle swarm optimization algorithm to perform the inversion. This method is compared to inversions by support vector machines, which are now common for defect characterization. The proposed inversion method appears to be an useful tool for decision analysis, highlighting the more likely results. All such elements are appraised from both synthetically generated and laboratory-controlled experimental data, which illustrates that a fast and efficient method has been indeed developed.