This might be useful, despite thread necromancy from 2008.
Sigma is the Standard Deviation for a Normal curve. For any given normal curve (shown in that linked picture above) there is a standard deviation where ~34% either side of the mean (peak) will fall.
There's a great picture on the wikipedia page:
http://en.wikipedia.org/wiki/Normal_distribution#Standardizing_normal_random_variables.
Normal random vs ... typical random.
If you drew a normal random from 100, you are equally as likely to get 2 or 98 (each is 1/100 chance) as you are to get 45, 50, 60, 73 etc. With a normal curve, the further you are from the mean (50), the lower the chance of that number appearing. So, 2 or 98 will occur very infrequently (~1%) and most results will land around the 50 mark. If your sigma is larger, then the curve is flatter and in the 1-100 example, there will be a greater chance to get numbers between 30 and 70.
This is a really cool feature, although I haven't actually found a use for it yet.