I was led here by a link in a more recent, similar question, but I can't let the wavelet-based (Noise Ninja, Nik Dfine, Imagenomic Noiseware, Lightroom's own NR) answer stand alone without competition. Another product worth looking at is Topaz Labs' DeNoise (currently in version 5), which is not wavelet-based and can remove a lot of noise while retaining a lot of detail.
I have mentioned this product in a number of other threads, and was turned onto it by this answer posted by John Cavan some time ago. There is also this review at Luminous Landscape to take a look at. I've been able to nearly eliminate noise altogether while retaining all of the relevant detail, and actually had to back off a bit and use the "add grain" tool because the areas of flat colour (like, say, a clear blue sky) looked Photoshopped in with the gradient or bucket tools. I can understand how it can recognize longer edges, but there is some deep voodoo in how it distinguishes subject texture from image noise.
That said, it isn't perfect either -- the kinds of noise generated by early-vintage CCD sensors (as opposed to CMOS) at the highest ISOs seems to be immune from non-destructive noise reduction no matter what algorithm is used. (That noise, if you've never seen it, looks rather like rough and careless cross-stitching. Individual pixels don't go rogue; rather, short runs of rows and columns of pixels will shift luminosity or colour dramatically as a group.) And converting to B&W doesn't help much either in that case -- the only real answer is to downsize the image and be happy with an overgrown thumbnail you can share on the web.