i just ran a check with the sound card:

a) the sound card contributes noise. i shoved in a 50 ohms resistor
into the mic input, cranked up the mic level, save a 'silent' wave
file and saw it through a hex editor. There is a random noise of about
2-3 bits. That means, the effective range is not 16 bits but 12 bits.
12 bits can represent 4096-1 as the highest number. That means, the
dynamic range is 4096 squared: about 16 million or 72db. A little low
really for interesting work. there isnt enough 'roof' to resolve a
spur about 60db down without touching the grass

b)the sampling is at 44khz. some of the posts got confused between
20khz b/w and 20khz sampling rate.

hans is right about the PC being a kludge when compared to an
oscilloscope. I live in India and second-hand oscilloscopes are a
rarity. The new ones cost an engineer's whole year's salary. I have
purchased a tektronix 454 last month on ebay for $300, it is still on
its way to India. So, while i can imagine that in a number of more
developed countries oscilloscopes are not really a problem, they do
remain a problem in many parts of the world.

More often than not, an amateur already has a PC. Adding a simple
hardware to be able to do quick and dirty spectrum analysis might be
an interesting option. As Wes writes in SSD and EMRFD, the purpose of
test instruments is to help with the projects on hand, rather than be
projects themselves (then he went on to homebrewing a spectrum
analyser, hehe).

- farhan