ITechniques for the efficient derivation of decision trees for very large data sets are summarised. When the semantics of the variable is such that an order operation is meaningful and the possible number of values is considerably less than the number of examples, it is shown that the consideration of single value tests at an internal node can proceed much faster than the more standard assumption of a continuous variable. The computational complexity of the most efficient continuous test is compared with the most efficient binary test on such ordered variables. Two simple data sets with a relatively moderate number of examples is used to time the derivation of decision trees with the various techniques. It is shown that the trees with a simple binary ordered test tend to be more effective than more general tests at the nodes.