In Eddy-Current Testing (ECT), we are interested in obtaining information about the flaws possibly present in the inspected parts. Choice of model is a general problem that one can run into when analyzing the shapes, the number of flaws and furthermore the influence of nuisance parameters, like the lift-off. In the framework of Bayesian inference, we need to calculate the evidences of the concerned models in order to choose between (or among) models. Nested Sampling (NS) shows the possibility of approximating the evidence with reasonable computational cost. This contribution proposes a model choice method based on an improved NS algorithm the aim of which is to get independent samples with hard constraint on the likelihood value in a more efficient way. It works for models who have Gaussian-like likelihood distributions. By analyzing the model evidences approximated by the NS algorithm and the final active samples, this method makes it possible to assign the correct model for the flaw of concern and meanwhile to estimate the corresponding flaw parameters. Simulations have been conducted to validate this method. The results confirm its computational efficiency and model choice reliability insisting the fact that metamodels contribute to the efficiency in the most complex cases.