Hi. I would just like to present a viewpoint which I would never have thought of as controversial before I came to this board: If you can record without overloading anything, the ideal in fact is to get your signals to peak as close to 0 dB as you can get without actually reaching 0 dB--rather than the -12 dB that several people have mentioned in this thread.
It's one thing to aim for a higher peak level and end up with your peaks down at -12; that can happen, and for a hobbyist recording it's usually no tragedy. That's because at this point in the development of audio technology, our recording equipment has a considerably wider dynamic range than much of the material that many of us have the opportunity to record.
For example, if you're in a venue with a noise floor of 40 dB SPL and the peak SPL of the program material is 110 dB, then the difference is 70 dB. So it only takes 80 dB dynamic range (10 dB greater than the distance from the noise floor to the peak SPL) to capture that sound with no audible noise contribution whatsoever from the recording equipment. In that example situation, a peak level of -12 dB, even on a 16-bit recorder with a dynamic range of 90+ dB, would be adequate (and incidentally there would be no advantage whatsoever in recording 24-bit in this example, either).
But assuming that you have competently designed equipment that is appropriately chosen in terms of sensitivity and overload margins and used correctly, nothing whatsoever is gained by setting the peak recording levels low on purpose. All that's happening is that (in the example I gave) you can afford to throw away those 12 dB because the full dynamic range of 16-bit recording is enormous, and because the venue is as noisy as it is.
In other situations, however, you would not be able to afford it, i.e. some of those 12 dB will turn up as audible noise in the recording.
--best regards