Approximation by Penultimate Extreme Value Distributions

n certain cases the distribution of the normalized maximum of a sample can be better approximated by a sequence of different extreme value distributions than by the final one. We show that these cases are rather restricted and that the possible improvement is not spectacular.