I’m not usually much of a “visual” guy. I’d rather read a text description of most stuff than look at pictures.
The one great exception to this is when it comes to modelling large databases–say, more than 10 tables. At that point, staring at line after line of SQL simply doesn’t cut it–I can’t discern the problems at a glance, there’s lots of paging around, etc.
For the moment I use the UML mode of “Dia”:http://www.lysator.liu.se/~alla/dia/, which is an OK tool, plus a custom XSL stylesheet for coverting the Dia’s output into SQL. It works OK–you do get a diagram out of it, and you can coerce it to produce decent SQL–but it’s just not…fluid. There’s a lot of fiddly stuff that requires very careful work, and supporting things like foreign key references always requires more work than it should. This makes creating the diagram that much more work and frustration.