The past demographic history of a population or a group of population has a strong effect on its nuclear genetic diversity, which implies that current patterns of molecular diversity could be used to reconstruct his past history. A wealth of statistical techniques has been developed to that aim, but most of them are restricted to relatively simple scenarios. Quite recently, an Approximate Bayesian Computation (ABC) framework has been developed to estimate parameters of more complex evolutionary scenarios. This approach does not require the computation of likelihoods, but rather compares observed and simulated data by means of summary statistics, and tries to find sets of parameters minimizing the distance between observed and simulated summary statistics. This quite flexible approach can be basically applied to any model that can be simulated, but it requires very large computational power and it becomes quite tedious when the time taken by the simulations and/or the computations of summary statistics is long. It would be therefore good to find ways to reduce the number of simulations to be performed to infer parameters, which is the object of this line of research.