h2g2 The Hitchhiker's Guide to the Galaxy: Earth Edition

Picking the numbers is in fact a science. The vast majority of players will assume that any set of numbers is ans good as any other and will pick the age of their dog, their father's birthday etc. This leaves the scientists to consider which numbers are the most popular and to pick those numbers that are least likely to be selected by any random member of the public, since although no one set of numbers is more likely to be drawn than any other, if you win you may as well win big by being the only person with the winning combination.

This is the reason why lottery systems allow a random selection by machine to be used since this increases the chances of large wins (by selecting unpopular combinations). Otherwise there is a danger that the outcomes would tend to be no winners or multiple winners (all winning less than the vital million). If you recall, Camalot had the lottery rules changed so that in the unlikely event of the 3 ball matches wiping out the jackpot, that they wouldn't be liable to cough up a tenner to everyone. This suggests to me that some combinations of numbers were (are?) in danger of crossing this threshold if they had been drawn.

The parallel computing group at University of Southampton even analysed the history of draws and prize breakdowns to try and work out what combinations of numbers were popular to aid the selection of an unpopular bunch of numbers.

In summary I think the rules of thumb are:
-choose big numbers (over 31 is good)
-don't go for all primes or you will find yourself sharing the jackpot with 150 maths teachers.
-choose short runs of consecutive numbers, a run of 2 or 3.
-use dice/cards/rand() or whatever source of entropy you believe in, but if it looks like a non-random set when you write it down, or when you fill in the lottery card, reject it and try again.