To my knowledge, no one has written down any guidelines.

I segment my models on 8.5 x 11 pages, usually landscaped. By the time I sit down with the tool, I have already done many model fragments on whiteboards and note paper so I have a good idea as to the general structure of the "important facts". These involve key system object types and more or less map to (in ER terms) major entity types. I then decide which to present first, second, and so on in a "top down fashion" so it makes some sense as you read the model from front to back and a given page doesn't use a major, complex object type not already (somewhat) defined before it.

If a given object type has many facts about it that map to attributes (they have a exactly one or zero to one relationship to the OT) I lay those out more or less arbitrarily with the OT in the middle and the related facts down each side. I often re-arrange fact types grouped by contestant participation: e.g. if three facts are in an external uniqueness constraint, they lay next to each other.

Facts about "lookup tables" (e.g. Thing Type Code has thing Type Name) and audit information (e.g. Customer has insert Date) I stuff in the back pages of the model.

Hope that helps,
Scot.