The definition and use of boundary cases is a common approach when aiming to anthropometrically accommodate a desired percentage of the targeted population by a design. The cases are defined based on anthropometric data that represents the targeted population. Approaches that define cases based on the variation within just one body measurement are poor for most design problems in representing anthropometric diversity. Hence, the consideration of variation within several body measurements is preferred. However, an approach that is based on performing several separate studies of the variation within a number of measurements leads to undesired reduction of accommodation due to the lack of consideration of the effects of correlations between measurements. This paper compares theoretical accommodation levels when using an ellipsoid versus a cuboid based approach for defining boundary cases to represent anthropometric variation within three body measurements. The ellipsoid approach considers correlations between body measurements whereas the cuboid approach does not consider correlations between body measurements. The paper suggests the application of the ellipsoid method for defining boundary cases for better reaching desired accommodation levels in boundary case based design problems. These cases can be used to define computer manikins when using digital human modelling tools. The method is also applicable when wishing to select extreme but representative real people to be involved in physical fitting trials.