Abstract : Virtual reality enables users to manipulate virtual objects from local or distant sites. Nevertheless, virtual objects are usually manipulated by only one user at a time. In this thesis, we define an interaction protocol describing how interaction tools and interactive objects communicate. This message-based protocol handles local and distant interactions on interactive objects for single or multi-user situations. It also manages permissions for access of interaction tools. Moreover the development of this protocol led to the definition of interaction tools and interactive objects. This characterization has driven an implementation based on a set of reusable software components that let an interaction tool modify how it selects or modifies interactive objects, and let an interactive object modify its own behaviour when it is manipulated. On top of that, extensions to the COLLADA file format based on this characterization and dedicated to collaborative interaction are proposed. A new technique for multi-user interaction is proposed. Movements in 6 degrees of freedom are deduced from the positions (not orientations) provided by 3 virtual hands. This technique thus enables two or three users to manipulate together virtual objects. The ability given to users to put precisely their virtual hands brings them a way to imitate better reality. In fact, users' feeling during a manipulation can depend on how users hold this object notably with cumbersome objects. Therefore, this technique aims to obtain realistic hand positions and gestures of users. The implementation is built upon the interaction protocol. A new concept of reconfigurable tangible interface for single or multi-user interaction is also proposed. This interface provides a way to manipulate virtual 3D objects. This interface, named RTD, proposes a reconfigurable physical mesh made up of manipulation points that have to be put on a virtual object, like grasping points. This mesh thus roughly sketches the form of the virtual object to manipulate. Two kinds of RTDs are illustrated in this thesis. A first kind of RTD is a triangular shape. It is named RTD-3 and so made up of three manipulation points. Experiments show that users prefer this technique over classical techniques of virtual reality like the mean of users' movements or the separation of degrees of freedom. A second kind of RTD is made up of four manipulation points and exists in two versions. The plane RTD-4 has its four points laying on a plane to create a quadrilateral. The non-plane RTD-4 uses articulated joints at its manipulation points to obtain a tetrahedron. A 3D form for a RTD would lead to better matching between the form of a RTD and the form of the virtual object to manipulate.