Three different imputation methods are performed: two based in the closest neighbor inside a stratum (donors methods) and the third based on Bayesian models. Comparations are carried out under 3 different scenarios of missigness and relative error is used to quantify the quality of the imputations. According to our results, donors methods are simple and useful but, once we have high missing rates (20%), they give problems. Bayesian models, richer but more difficult to interpret, are more insensitive to bad missing conditions. New strata are proposed for Sd2012 via Bayesian models interpretation and perspectives of updating Bayesian models with Sd2012 data are explained. . Indirect sampling is used in hard-to-reach-populations. Target and sampled populations differ and knowledge about the links between both is required. Imputation problems arise when values of links are missing. In the framework of the "Sans domicile" survey SD2001, different methods have to be proposed and compared