Abstract. This paper presents a method for analogizing high-dimensional spacetime curves, and shows how it can be used to transform a motion sequence into new content and styles according to two accompanied reference sequences. By providing examples, our system can estimate the parameters of a statistical model which incorporates spatial and temporal factors for motion curves. A user can then easily modify existing motion sequences, within these parameters, to have new semantics as examples. This method can also be used to enrich motion capture databases or edit recorded animations in batch mode.