This communication addresses the problem of any simulation tool: the accurate and efficient sampling of a physical observable with respect to a parameter. A popular sampling technique is the uniform sampling combined with a straight-line interpolation for representing the continuous variation of the observable. However, this sampling becomes rapidly inefficient if the observable varies strongly since a high-oversampling is necessary du e to Nyquist's theorem. An alternative is nonlinear sampling and nonlinear interpolation of the sampling points. Another reaso n why more efficient sampling techniques are needed is the optimization of devices using full-wave simulation tools where the reduction of sampling points is essential to accelerate the design of a component. This paper presents an algorithm that is ba sed on ideas coming from the model-based parameter estimation (MBPE) and the Genetic Algorithm (GA).