Michelle Jellinghaus <michelle@EMODE.COM> replied to her own problem with this code: > In the event that this may be useful to others, I have included the > beautifully simple solution below. It seems to work well for my purposes :) > > proc sql; > connect to oracle(user='xxxxx' pass='xxxxx' path='xxxxx'); > create table mysample as > select * from connection to oracle > (select * > from mytable sample(.01) A1 > ) > ; > quit; > > note: this query returns 0.01% of the records in this table, in random > order.

Very nice. And let me say that I really like the way that you structure your code for readability, which too many people forget. But...

Does it provide a simple random sample With Replacement, or a simple random sample Without Replacement? This may matter, depending on your usage.

Does it provide a repeatable process, so that you can substantiate your work, or reproduce your sample when the big boss questions your results?

If these things don't matter, then you have a great solution. If some aspects of the sampling protocols do matter, then you need to think about some of these details.

HTH, David -- David Cassell, CSC Cassell.David@epa.gov Senior computing specialist mathematical statistician