Rick
For some of the nitty-gritty details check out the
JLS section 5.1.2 This will make frequent reference to the IEEE 754 standard - which I can't find openly published on line but if you want a good article on it check out
Suns numerical computation guide.
As you can see it is pretty intense - at least for me it is
The upshot of it is (at least form almost everyday programming) that you can lose precision with some conversions (both widening and narrowing). It is most important to know that it does happen and why, but the actual math involved is usually not required learning - unless you're actually creating a compiler or high precision program of some sort. If that is the case you can always try to find one that supports float-extended-exponent value set.
hope that helps