An inverse problem is analyzed where corrosion of rebars is detected from a small number of potential data measured at the surface of concrete structure. Because the shape and number of corrosion in rebars are not known in advance, usual inverse analysis method in which the shape and number of corroded part are assumed is not available. In this research, the genetic algorithm (GA) is employed without any assumption. The fitness in the multi-step GA is defined as the inverse of difference between experimental and numerical potential values, and is evaluated by the boundary element method (BEM). To reduce the computational time, the net elements, which have been recently developed by the authors for corrosion analysis of net structures, is used together with the multi-step GA. It is shown by a simulation that the multi-step GA with net elements are successfully employed in the inverse analysis.