I encourage them to create a character without mechanics first and then do the character sheet last.
If that's the order, then I'd recommend point buy (for 3e, 28 or 32 points, depending on campaign power level). Or, if you trust your players not to abuse it, just let them select their stats.
I've tried a whole bunch of creation methods over time, and have found, on balance, that point buy tends to give the most satisfying results (despite the min-maxing that seems to be an inevitable consequence). I've found that random rolls work well in theory, but always seem to end up with one PC having much better stats than the rest (and usually one with much worse stats as well).
(That said, the standard 4d6-drop-lowest does seem to work reasonably well, most of the time.)
One method I really like (but which is totally unsuited to your campaign) works as follows: The players roll 4d6 (drop lowest)
seven times, and keep the results in order. They may then swap any two numbers around, and then drop one stat from the set, but have to use the resulting numbers in order. (The 'swap' and 'drop' steps can be done in either order.)
(So, a player might roll 9, 12, 15, 16, 8, 10 and 7. He can then swap any two numbers (for example, to get 16, 12, 15, 9, 8, 10, 7), and then drop one (to get 16, 12, 15, 9, 10, 7), but then has to use the numbers in order: Str 16, Dex 12, Con 15, Int 9, Wis 10, Cha 7.)
However, the inability to order the stats as desired makes this method unsuitable for your use: the players are really going to need more control over which stat goes where to match up the mechanics of their characters with the concept.
Another method I experimented with, but ultimately discarded, was to use a set of playing cards with all the 7's though 10's and all the 'face cards' removed (leaving only the 1's through 6's). Each player would then shuffle the remaining cards and deal six hands of four cards each. In each hand you then discard the lowest value card, sum the rest, and arrange the numbers as desired.
This method gave some element of randomness in the stats, while generally eliminating the "one player rolls really high; one rolls really low" issue from the standard 4d6. However, I generally found that it left most characters with one really low stat, moreso than occurred with the standard 4d6.