This report describes a system designed and implemented to select subsets of "interesting" features from a planar motion sequence of a human stick figure. This system continually determines interesting features by segmenting the performer's space according to a center of attention, evaluating feature interest, and updating the attentional center. Feature interest is evaluated using thresholded Euclidean distances in a feature-interest space that incorporates three factors: current movement, captivation, and task-level bias.