Eddy-current testing (ECT) is a widely used nondestructive evaluation technique. The numerical simulation of ECT methods involves high complexity and computational load. However, one needs reliable solutions (within a reasonable CPU time) for these problems to be able to solve the related inverse problem. One possible approach is to build a configuration-specific database, consisting of well-chosen samples (corresponding input data -- output signal pairs). Once the database has been constructed, the sought information can be retrieved practically in no time. However, the optimal choice of samples arises complex optimization problems. This paper presents a sampling method which aims to achieve databases being optimal in a certain sense. The goal of our approach is to spread out the output samples in the whole conceivable output domain. The method is formalized as a maximin problem which is solved step-by-step using the kriging prediction.