The realization of a direct approach to an inverse problem on a parallel MIMD computer system is considered. On each step of the Levenberg-Marquardt minimization algorithm an effective iterative process for solving direct scattering problems based on fictitious component methods and iterative methods in subspaces is used. Parallelization properties of both outer and inner iterative methods lead to a good speedup.