>Sergey,
>
>I have seen this mentioned in some SQL tips, but in "official" docs I had not seen
>stated that NewID produces random data
>(in the statistical meaning - usable for sampling&testing)
>or the scale level it can be used at (more than ordinal ?)
>
>Did you read some MS doc certifying fitness for that usage
>or a test on the robustness when used this way ?
>Any links to throw my way ?
>
>thx
>
>thomas
>
>
I've seen a very good link where this concept was used by Paul White

Advanced T-SQL Tuning - Why Internals Knowledge Matters

BTW, you can also check TABLESAMPLE in BOL (it does show NewID() approach).