Linearization is one the very usual technique used in statistics to approximate and estimate the variances. A large set of publications are dedicated to methods of linearization including linearization of a function of totals, of estimating equations of nonlinear functions of the observations. Several techniques have been proposed. First we present a synthesis of these different methods. Next we derive a new method consisting in differentiating by the sampling indicators. This method can also be applied to calibrated estimator. Finally, we show that this technique of linearization can also be used to take the variance due to the nonresponse in reweighed of imputed estimators into account.