This thesis proposes a computational aesthetics methodology for measuring the design quality of poses in animation data, and then for predicting the composition skill of the source artist. We draw from animation and performing arts principles to select pose features and design metrics for supervised learning on a corpus of extracted 3D poses. Though our approach is designed to be general enough to apply to aesthetic features in any performative figure data, we choose three features to investigate and conduct specific experiments using motion captured data from live performers in the domain of dance. An initial pilot study is conducted on pose data from a dance instructor to assess our metrics, and then a formal experiment is conducted on performance captured data from participants playing a popular Kinect dance videogame. Principal component analysis is utilized to identify low-level skeletal features, and then supervised learning experiments are conducted to infer performer skill from figure composition quality based on metric scores. Results show statistical correlations between intuitive skill rankings, game score distributions, and metric ratings. This thesis provides a methodological foundation for future work in scientifically studying the arts to formalize principles of figure representation.