Dcrigger is absolutely right. If you're using a 16-bit system, keep the headroom. If you can change your system to 24-bit, then you can afford less headroom. I have some software that runs at 32-bit, but that's not particularly commonplace.

The number of bits determines the total dynamic range available to you. If you imagine the amplitude (volume) as 'steps' then a 16-bit system has fewer 'steps' to the top than a 24-bit system. They get to the same level ultimately, but the steps in 16-bit are larger and hence less accurate. If you want to get even more technical, you also have the 'sampling rate' which determines the number of samples of sound taken per second.

If you imagine sound as a wave, a computer can't actually process the entire wave. So, instead it takes a 'sample' of the wave at defined periods. By something called the Nyquist-Shannon theory, the number of times a sound is sampled a second is approximately double that of the maximum frequency produced on playback. If you assume that the maximum frequency Humans can hear is 20KHz (which is what is stated, although it's usually lower) then you need to sample the signal at 40KHz - i.e. 40,000 times a second. In practice, there is some space allowed, so the lowest modern sampling rate is usually 44.1KHz. For things like phone services, the encoding is actually much poorer, but that's to save bandwidth and isn't necessary for the purpose.

A CD therefore, is encoded at 16-bit and 44.1KHz sampling rate. Which is adequate for good replication of sound. There are now much better technologies (including DVD and Blu-Ray) that have much higher theoretical bit-rates and sampling rates.

This is all basic digital audio theory. If you have questions, feel free to ask. This is part of my degree.