The Capture-Recapture method was developed in animal ecology to estimate animal population parameters. Its applications in epidemiology, developed at the end of the 80's, set specific problems and generally lead to over-estimations of the sizes of the populations, due to the fact that the sampling lists are not independent and do not cover the same population. In this context, we propose an adaptation of the ``Robust Design'' to epidemiology. This method was proposed in zoology to permit a robust estimation of the population parameters. We obtain a new estimator that allows us to relax the assumption that the various lists sample the exact same population. We compare it to the classic estimator in different cases, in terms of bias and asymptotic variance, using simulations for the latter. The new estimator has various qualities (bias reduction, robustness with respect to lists sizes, efficiency) compared to the classic estimator, in all the considered cases.