A natural and intuitive method is proposed to help a user manipulate an object in a virtual environment. The method does not need to assign special properties to the object faces in advance and does not require special hardware. Instead, it uses only the visual constraints of motion among object faces that are dynamically selected by a real-time collision detection method while the user manipulates the object. By constraining more than two faces during the user's manipulation, the proposed method provides an efficient tool for complicated manipulation tasks. First, the method of manipulation aid is described. Then several experiments demonstrate the effectiveness of this method, particularly when the user is requested to precisely place a virtual object in a certain location. Finally, as an application of the proposed manipulation aid, an experiment is conducted to compare the performances of a task (constructing a simple toy) in a real versus a virtual environment. Results show that the distance accuracy and completion time of the virtual task with the manipulation aid is close to that of the real task.