A new mean of solution of eddy current testing (ECT) inverse problems is presented herein. The aim is to characterize a defect within a generic work piece by inverting an ECT signal in a short time (less than 5 s on a standard PC). The inverse problem is solved using Particle Swarm Optimization (PSO). As inputs of this algorithm, a metamodel is used in order to reduce the computation load and to contribute to building up a fast ECT inverse method. It is tested on a number of synthetically generated data, the chosen work piece being a tube. It is also tested on two sets of measured data. The proposed method shows good accuracy, satisfactory speed, and provides additional information besides the inversion results, which highlights possibly indeterminate cases. So it is useful as well for decision analysis