>>According to BOL: "We recommend that you not use time-dependent functions, such as RAND(), within a NULLIF function. This could cause the function to be evaluted twice and to return different results from the two invocations".
>>My guess is the value that NULLIF() checks is different from the value returned because of that.
>
>Sergey,
>where you found this?
>I read all about RAND() and NULLIF() bud I didn't read anything like that.

I found it in the NULLIF() topic (Remarks) in BOL 2005 online at http://msdn2.microsoft.com/en-us/library/ms177562.aspx. It also present in BOL 2005 on my PC. It's not in BOL 2000, though.