The first way to generate a random number is by using the SQL Server RAND function. This will return a float value.
Both of the techniques Chad shows are examples of generating uniform distributions—distributions in which any value is just as likely as any other. There are plenty of places in which a uniform is great: drawing by lot is one of them. But when you’re generating artificial data, the results tend to look unrealistic because not many natural phenomena follow uniform distributions.
If you’re interested in generating numbers which tend to look more realistic when generating artificial data, I have a post on generating random numbers built on a normal distribution.