What methods are there for selecting an "appropriate" cutoff height in a dendrogram when performing hierarchical clustering on sample data points to partition a hyperspace into subspaces?

Little bit of background:

I'm looking at using the CHAMELEON hierarchical clustering algorithm to create a dendrogram of sample data points generated by a procedure that imitates an arbitrary problem space. I'm looking for a way to choose an "appropriate" height in the dendrogram in order to partition the problem space in an "intuitive way." Does anyone know of an algorithm for doing this? I've searched, but either such an algorithm doesn't exist (which I highly doubt) or I'm choosing the wrong words in my searches. If it makes any difference, the clusters will probably end up being hyper-ellipsoidal.