Accurate self-assessment of body shape and size plays a key role in the prevention, diagnosis, and treatment of both obesity and eating disorders. These chronic conditions cause significant health problems, reduced quality of life, and represent a major problem for health services. Variation in body shape depends on two aspects of composition: adiposity and muscularity. However, most self-assessment tools are unidimensional. They depict variation in adiposity only, typically quantified by the body mass index. This can lead to substantial, and clinically meaningful, errors in estimates of body shape and size. To solve this problem, we detail a method of creating biometrically valid body stimuli. We obtained high-resolution 3D body shape scans and composition measures from 397 volunteers (aged 18–45 years) and produced a statistical mapping between the two. This allowed us to create 3D computer-generated models of bodies, correctly calibrated for body composition (i.e., muscularity and adiposity). We show how these stimuli, whose shape changes are based on change in composition in two dimensions, can be used to match the body size and shape participants believe themselves to have, to the stimulus they see. We also show how multivariate multiple regression can be used to model shape change predicted by these 2D outcomes, so that participants’ choices can be explained by their measured body composition together with other psychometric variables. Together, this approach should substantially improve the accuracy and precision with which self-assessments of body size and shape can be made in obese individuals and those suffering from eating disorders.