> Do you want equally likely labelled trees or equally likely > unlabelled ones? For the labelled trees, the method is fairly > easy; there are 2^(n-1)! labelled trees with n nodes, and going > from n-1 to n can be done by selecting a node at random and > branching there. For unlabelled trees, the problem is much > harder.
Unlabelled. BTW, I forgot to mention that I am mainly interested in full binary trees (each internal node has exactly two children).