When hand shape, hand position and hand orientation are changing simultaneously such as in the physical manipulation of objects shape, the dimensionality of the control is highest. Although the manipulation of shape (i.e. claying or moulding) in the physical world is most effective with touch and force feedback, our assumption is that these forms of feedback can be replaced by acoustic feedback with some compromises. The generation of appropriate touch and force feedback, while exploiting the maximum range of motion of gestures, is currently technically too challenging in a shape manipulation task. We are also interested in the use of dynamic signs (hand shape is constant, but hand position and/or orientation is changing), although they represent less dimensions of simultaneous control. The other end of the spectrum, in terms of control dimensionality, is represented by static signs, which are not useful in the task we are interested in (i.e. multidimensional control) other than for selection. In previous work on gesture interfaces such as [2] it has been noted that, since humans do not reproduce their gestures very precisely, natural gesture recognition is rarely sufficiently accurate due to classification errors and segmentation ambiguity. Only when gestures are produced according to well-defined formalisms, such as in sign language gesture recognition, will automatic recognition have acceptable precision and accuracy. However, a gesture formalism will require tedious learning by the user. Thus we do not compute or analyse any abstract symbolic representation of the gestures produced by the user, but instead focus on the continuous changes represented by the gestures.