Abstract : Over the past decade, many elds of discovery have begun to use motion capture data, leading to an exponential growth in the size of motion databases. Querying, indexing and retrieving motion capture data has thus become a crucial problem for the accessibility and usability of such databases. Our aim is to make this approach feasible for virtual agents signing in French Sign Language, taking into account the semantic information implicitly contained in sign language data.We propose a new methodology for accessing our database, by simultaneously using both a semantic and a captured-motion database, with dierent ways of index- ing the two database parts. This approach is used to eectively retrieve stored motions for the purposes of producing real-time sign language an- imations. The complete process and its in-use eciency are described, from querying motion in the semantic database to computing transitory segments between signs, and producing animations of a realistic virtual character.