Microsimulations may involve a large number of agents. It is then practically impossible or too expensive to obtain a fully and complete disaggregated data set about these agents of interest. Moreover, if such a dataset was available, its use would be potentially problematic in view of stringent privacy laws. To address this problem one may build an articial population starting from known aggregate data. Most of the known generation methods are explained in this paper. Their advantages and limitations are discussed and references are given for further details.