Abstract: Realistic and directable humanlike characters are an ongoing goal in animation. Motion graph data structures hold much promise for achieving this goal; however, the quality of the results obtainable from a motion graph may not be easy to predict from its input motion clips. This article describes a method for using task-based metrics to evaluate the capability of a motion graph to create the set of animations required by a particular application. We examine this capability for typical motion graphs across a range of tasks and environments. We find that motion graph capability degrades rapidly with increases in the complexity of the target environment or required tasks, and that addressing deficiencies in a brute-force manner tends to lead to large, unwieldy motion graphs. The results of this method can be used to evaluate the extent to which a motion graph will fulfill the requirements of a particular application, lessening the risk of the data structure performing poorly at an inopportune moment. The method can also be used to characterize the deficiencies of motion graphs whose performance will not be sufficient, and to evaluate the relative effectiveness of different options for improving those motion graphs.