But this does not really matter, because the distribution of this approach is not even. If you have ids 1,2 and 4 then ‘2′ is twice as likely to be chosen. This is the reason I haven’t mentioned this approach – it is not random.

There are up to 10 different approaches, and they all suck one way or the other. :)

Another one is:
select count(*) from mytable
$rand=rand(1,numRec)
select * from mytable order by id limit $rand,1

BUT please don’t use this. Yes you can use ‘where’ with it. Yes you can have holes. Yes it is simple. Yes you can get more IDs at once, if you build a function that returns N different randoms from X values. Yes the distribution is even.
But it is also (probably) terribly inefficient. In worst case it is probably as inefficient as rand(). If you must take this approach, at least benchmark it. i haven’t done it because… Well, I think it is inefficient. :) But yes, MySQL could optimize that kind of query, so YMMV.

Also, your solution only retrieves 1 record from DB, not multiple. It can be changed to do so, though.

But not much can help you if you have a where constraint:
select * from mytable where somefile=3 and someotherfield=5 order by rand() limit 3;

I am still looking for a solution to that case. Currently I am assigning a rand() every few minutes to some field and selecting like this:
$rand=rand(1,99999999)’;
select * from mytable where somefield=3 and someotherfield=5 and randfield>=$rand order by randfield limit 1

If anyone has a better idea… Do share. :)

Here is the bug report, let’s hope MySQL developers step up and solve this issue for us so we don’t have to reinvent this functionality:
http://bugs.mysql.com/bug.php?id=33837

You can help by voting for it (hint ;).