Dealing With Overfitting

Either stop growing the tree earlier or prune it after-wards. Pruning has been more effective.
Use a separate set of examples (not training) to evaluate the utility of post-pruning nodes.
Use a statistical test to estimate whether expanding a node is likely to improve performance beyond the training set.
Use explicit measure of the complexity for encoding the training examples and the decision tree. Stop when this encoding size is minimize. Minimum Description Length principle (later).