This paper presents a global strategy aiming at solving efficiently inverse problems classically faced in the community of non-destructive testing, like flaw characterization and sensors optimization, for instance. The approach is based on intensive use of simulation tools that are obtained from complex physical models developed at CEA LIST and dedicated to ultrasonic and electromagnetic testing applications. Results presented in different physical contexts show not only its efficiency to solve parametric estimation problems, but also its capability to account for variability due to uncertainty in some model parameters and to evaluate the so-called well-posedness of inverse problems considered.