The paper presents the law-smooth update scheme, a new smoothed scheme for law updating in order to improve the optimization performance of the cross-entropy algorithm, particularly for large spaces and difficult functions. The method deals with problems related to the smoothed updating of different families of distributions, including Gaussian distributions, and to conditioning. It offers a clear improvement in terms of robustness to convergence degeneracy. Comparative tests on difficult problems show that this new approach considerably improves convergence rates compared to the classic update scheme. An application is presented on the minimax optimization of target localisation with IR camera by the Cramer-Rao bound.