[Siegmund]

All you have to do is define a measure of goodness, then see how each of the hands score on it. There is no one unique measurement - and no one unique ranking of starting hands, either! AA is best on all the lists, performing better against just about any lineup of hands you throw at it than any other does.

One possibility is to calculate your chance of winning a heads-up all-in vs one random hand. If so, 32o is the worst hand, not 72o. Another possibility is to calculate your chance vs. 9 random hands if they all see the showdown; this overvalues suited connectors and undervalues things like TT. You could observe a large number of hands, and keep track of how often each one won a showdown - but this will undervalue the hands which are so good they often win before a showdown. Abdul did a famous study several years ago with Turbo Texas Holdem and calculated the EV of each initial action on each UTG starting hand vs. a certain set of AI opponents (who don't bet exactly the way real humans do.)
All those, of course, require a lot of calculating.

A few days ago someone proposed a definition for "hits the flop hard" and wanted to calculate the chance for each starting hand. Tedious, but tractable, and I think the result would have made a pretty good ranking list.

One thing you can calculate more easily than doing all possible hand pairs is calculate the chance that both hole cards will play for each starting hand. I made a list of that last summer; it wasn't bad but overvalued small pocket pairs badly (rating 22 about as good as KQs or AKo if memory serves).

So. Pick your definition of goodness, and crunch some numbers.