I think you are wrong. I think randomly drawing any significant number of people from a pool that isn't normally distributed is still likely to result in a normal distribution.
No, that's not quite right.
Randomly picking people from a distribution should eventually end up giving the original distribution.
Taking n people from a pool m times, then the means of those m samplings will be normally distributed however (given the constraints of the CLT of course).
http://en.wikipedia.org/wiki/Central_limit_theorem