If it is a workflow, then it is just a way to make a selection, like regular mask, except you separate based on contrast.
Briefly, what i do is to go the channel palette and check the R, G or B layers which have the best contrast that will help me make the selection (for e.g: selecting a human from a busy background). Doodle a bit to increase the contrast (channel mixer adjustment layer -> monochrome and pump the values of whichever RGB layers has the best contrast, play with blending modes etc), made amendments (painting in certain areas to help with the contrast) and there you have it - a mask!
This is an extremely brief account, but the bottomline is, you want to increase the contrast between your object and the background you want to separate so you can easily create a mask.
Most Channel books should have examples on this. Hope it helps.
Some point to ponder: dodging & burning is pretty much the same, except that you create the mask "on the fly" with your hands/cardboard/etc instead of making it photographically. The beauty of the image-derived mask is that it is very detailed and precise with minimum effort. Note that "unsharp masking" is closely related to this technique.
On the computer, there's also much more flexibility. E.g. you can use such image-derived masks masks for blending different layers into one picture, allowing you complete control (curves etc.) for each contributing layer which gives much better control than traditional masks.