Anthropometric data are widely used in the design of chairs, seats, and other furniture intended for seated use. These data are valuable for determining the overall height, width, and depth of a chair, but contain little information about body shape that can be used to choose appropriate contours for backrests. A new method is presented for statistical modeling of three-dimensional torso shape for use in designing chairs and seats. Laser-scan data from a large-scale civilian anthropometric survey were extracted and analyzed using principal component analysis. Multivariate regression was applied to predict the average body shape as a function of overall anthropometric variables. For optimization applications, the statistical model can be exercised to randomly sample the space of torso shapes for automated virtual fitting trials. This approach also facilitates trade-off analyses and other the application of other design decision-making methods. Although seating is the specific example here, the method is generally applicable to other designing for human variability situations in which applicable body contour data are available.