I've actually used both. If you think about it, using RAND(Checksum(NewID())) will only generate as many unique values as Checksum(NewID()) in any given run since Checksum(NewID()) is what generates the "unique seed".

So it comes down to what kind of numbers do you need? Is it better to work from a float and muliply it out, or start with a big number and make it smaller to fit your actual allowable range. Since they're both basically the same, I tned to work with what is fastest to generate (i.e. Jeff's method listed above).

By the way - RAND() isn't random at all. For example, try running this:

select rand(1)

go

select rand(1)

go

select rand(1)

go

select rand(1)

go

select rand(1)

go

select rand(1)

In other words - it generates a predictable result for the same seed, even on separate runs and on different sessions. The only thing that makes it "seem" random is that it will "pick" a seed during different runs if you don't pass it a seed.

It also apparently has something that only uses positive seeds , since

RAND(1) = RAND(-1)

So again, you really don't get anything other that somewhat slower perf out of using RAND().