A method is described in this paper for estimating, by means of stochastic simulation, the asymptotic variances of multipliers of nonlinear models. It is used to estimate the uncertainty of the results of eight policy experiments for a particular model.