This paper first exposes some of the defects in the interval arithmetic algorithms of Moore and Krückeberg. Then it identifies classes of problems on which these algorithms compute "optimum bounds".