In many ways, this investigation is a ``proof of concept'' that shows it is viable to consider sign language recognition: we are using 95 signs out of the Auslan dictionary [Joh89] of 4000 or so. Classification time of decision trees is approximately logarithmic in the number of training instances (because of the tree structure employed for classification) and similarly, classification time for instance-based learning techniques can be made logarithmic in the number of training instances, using k-d trees ([WK91]). Learning times for simple instance-based learning techniques is linear, and it has been shown that under certain circumstances, so is C4.5 ([Squ94]). If we have more samples, we can simply add them in, without the major penalty of a quadratic increase in the time-cost of learning. Furthermore, both systems can operate incrementally (i.e. more training instances can be added to the learner without significant time penalties), unlike neural networks, which much be retrained if new training instances need to be included.