We report on a series of experiments in which all decision trees
consistent with the training data are constructed. These experiments
were run to gain an understanding of the properties of the set of
consistent decision trees, and the factors that affect the error rate
of individual trees. The experiments were performed on a massively
parallel Maspar computer.  The results of the experimentation on two 
artificial and two real world problems indicate that for three of the 
four problems investigated, the smallest consistent decision trees 
tend to be less accurate than the average accuracy of those slightly 
larger.