Equate real-world (SPL), analog (dBu, dBV), and digital (dBFS).  Emphasise signal chain from reality -> analog -> digital -> analog -> reality

Real word:
	0 dB SPL	Threshold hearing
	20-30 dB SPL	Quiet room (listening environment)
	83 dB SPL	Nominal music performance level
	85 dB SPL	Potentially damaging levels begin
	103 dB SPL	Loud music peaks
	120 dB SPL	Maximum music level (loudest heavy metal concerts)
	130 dB SPL	Physical pain

Analog (nominal):
		Noise floor
	+4 dBu	Nominal level
		Increasing distortion approaching clipping point
		Clipping level

Digital:
	-144 dBFS	Noise floor (1 bit noise @ 24 bits/sample, undithered)
	-96 dBFS	Noise floor (1 bit noise @ 16 bits/sample, undithered)
	-20 dBFS	Nominal level
	0 dBFS		Maximum peak level

So, it's reasonable to think about how these levels might correspond.  ...

It's reasonable to consider the largest and smallest levels that will need to be captured and reproduced in order to give an exciting, authentic sonic experience.  The quietest sound we can hear is 0 dB SPL, and the loudest anyone would want to hear would be 130 dB SPL.  In practice, most listening rooms have a background noise level of 20-40 dB SPL, so there is little point in having a sound reproduction system with an inherent noise level (noise floor) much below that.  Furthermore, sound levels above 103 dB are very lound, uncomfortably loud, so there is little need to reproduce these.  So, while we would require a range of 120 dB to reproduce the full range of levels we can perceive, in practice, a range of only 60 dB (100 dB - 40 dB) would be acceptable.

Most kinds of music have a crest factor of around 20 dB, meaning that the difference between the nominal (average) level and the maximum (peak) level.  So, the nominal level in the recording system will be 20 dB below the maximum.  As it turns out, we tend to prefer the nominal level for listening to be 83 dB SPL (this is the level they calibrate film theatres to).  Therefore, the peak level will be 103 dB SPL.

The analog signal level you get from a microphone will be proportional to the sound level in the recording environment (the exact ratio will depend on the sensitivity of the mic). 


-- Analog -> A/D headroom

If you decide that +4 dBu (in the analog domain) will correspond to -20 dBFS (in the digital domain), then 0 dBFS = +24 dBu.  Most decent analog gear will provide this.  However, solid-state active components usually show increasing distortion before clipping actually occurs, so you need more headroom (about another 6 dB).  Thus, the clipping point of your pre-A/D analog chain should really be +30 dBu, so that the signal going into the A/D is clean even on peak transients.

-- Digital -> D/A headroom

For the digital-to-analog (output) process, the numbers are similar.  Again, 0 dBFS = +24 dBu (peak level), and -20 dBFS = +4 dBu (nominal level).  But again, we need a further 6 dB of headroom to avoid the analog distortion that happens as we approach clipping, so we have a +30 dBu clipping point required.  But wait: there's more!  As Bob Katz points out, it's possible for the output of a digital component to exceed 0 dBFS!  This happens...   So, your analog signal chain following the D/A converter really needs a clipping point of at least about +33 dBu, in order to cleanly reproduce the output from the D/A converter.