Quick theoretical explanation which might help some people in understanding what USM settings to try:
Sharpening was unnecessary in the days of analog output. A dot-based analog output device (such as a halftone screen) could make any shape of dot necessary to handle the detail structure of the image. If you had a hard edge going through a halftone dot, for example, the screening process would simply generate a pear-shaped or hourglass-shaped dot.
Digital output devices can't do that, because they're constrained to a fixed repertoire of output dot shapes -- whether they're the pixels of a monitor, the dither cells of an inkjet printer, or the predefined halftone dot shapes of a PostScript imagesetter. Sharpness inevitably would be lost when image edges passed through these dots, because the dot couldn't modify its shape. To compensate, digital imaging developers introduced various forms of sharpening, which simply exaggerates the edges of the image so they're bold and thick enough for a fixed dot pattern to reproduce them.
Unsharp masking exaggerates edges by heightening the contrast of the tones surrounding the edge; the closer to the edge they are, the more their contrast is increased. (This is why it's called "unsharp"; the selection mask that locates the edges is blurred so that the sharpening effect is applied more strongly closer to the edge. You actually can produce the effect yourself by using Photoshop's 'Find Edges' filter, saving the result as a selection channel, and then blurring it. You're doing exactly the same thing the Unsharp Mask filter does.)
Now that you know basically what's going on, you can understand what the three settings do.
-- Radius determines how far on either side of the edge the filter applies the sharpening; a low number focuses the effect tightly on the edge itself, while a high number adds contrast over a broader swath. Focusing the effect too tightly can prevent the sharpening effect from being visible; focusing it too broadly creates artificial-looking lines around the edge, like the heavy outlines around shapes in a comic-book illustration.
-- Amount simply controls how much the contrast is boosted near the edge. A higher amount creates a sharper-looking effect, but again, overdoing it will make the effect look artificial.
-- Threshold gives you some control over what the filter considers to be an edge that needs to be sharpened. Set it to 0 and the filter will think ANY difference in contrast is an edge that needs to be sharpened; the result is a drastic exaggeration of grain, noise, and other textures. Set it too high and the filter won't identify any edges that need to be sharpened at all.
Knowing this won't necessarily help you decide WHAT settings to use for a particular image being output on a particular device, but at least you'll understand what you're varying when you change the settings.
I think it's best to start by determining the radius -- it's the most important value, since it controls the width of the "track" along the edge (and remember from the start of this explanation, it's the thickness of this "track" that simulates sharpness by making the edge thick enough for the output device to render it.) How wide a track you need depends on the pixel resolution of your output device; if your printer resolves 240 pixels per inch, for example (good guess value for many inkjets) then the "track" won't be visible if it's much thinner than 1/240 of an inch at the final output size. If you want, you can figure out how many image pixels are represented by 1/200 of an inch or so of printer output, and use that as a starting point for your radius setting. Most of us, though, just use radius values that we've learned by experience give good results for various output devices and image sizes.
Once you've got a starting radius, you set the amount (remember, this controls how much the contrast is boosted along the edge) based on the structure of the image -- images with low "microcontrast" need higher amount settings. I always suspect that if I have to go over 150-160%, it's too much (although sometimes you need to do this to produce a specific effect.)
And finally you set the threshold so that you're applying sharpening to image details but not to artifacts such as noise and grain -- start with it at 0 and move it upward until these artifacts are excluded from the sharpening effect.
Addendum: I don't actually use USM much any more. I find that the "Smart Sharpen" filter in the newer versions of Photoshop is easier to use and gives more natural-looking results. If you've got it, try it!