Optimal ergonomic design for consumer goods (such as garments and furniture) cannot be perfectly realised because of imprecise interactions between products and human models. In this paper, we propose a new body classification method that integrates human skeleton features, expert experience, manual measurement methods, and statistical analysis (principal component analysis and K-means clustering). Taking the upper body of young males as an example, the proposed method enables the classification of upper bodies into a number of levels at three key body segments (the arm root [seven levels], the shoulder [five levels], and the torso [below the shoulder, eight levels]). From several experiments, we found that the proposed method can lead to more accurate results than the classical classification methods based on three-dimensional (3 D) human model and can provide semantic knowledge of human body shapes. This includes interpretations of the classification results at these three body segments and key feature point positions, as determined by skeleton features and expert experience. Quantitative analysis also demonstrates that the reconstruction errors satisfy the requirements of garment design and production.