I would usually start with 70% - 100% the width of lowercase /i and then eyeball it for fine tuning. There were several threads about this, here and over typophile.com (currently down), but I cannot seem to find them so easily...
Put a word that's loose on the right followed by a word that's loose on the left. See how tight you can go (at the low end of the intended point size range) before the words get too cozy. Then loosen it a hair.
I generally start with 250 (four-to-em) or 200 (five-to-em), as James suggested, and adjust to taste. And like Nick, I often use a layout app (InDesign) to test various increments to hone in on what seems optimum for targeted size and setting.
It depends on lots of factors, and you always need to test and refine by eye.
But also, if you measure lots of fonts and put the result in excel, you may get a Median result that looks something like this: For Sans: n advancewidth / 32 * 15(±2) For Serifs: n advancewidth / 32 * 13(±2)
For example: If you have a Sans and the /n advancewidth is 400 units, then 400/32*(15-2)= 162 (Smaller deviation) 400/32*15 = 187 (Median) 400/32*(15+2) = 212 (Biggest deviation)
if you have a Serif and the /n advancewidth is 400 units, then 400/32*(13-2)= 137 (Smaller deviation) 400/32*13 = 162 (Median) 400/32*(13+2) = 187 (Biggest deviation)
Pablo: Did you do that just with "regular" weight fonts? Per my comment above, within a family, as weight increases, usually advance widths go up and space widths go down.
Comments
But also, if you measure lots of fonts and put the result in excel, you may get a Median result that looks something like this:
For Sans: n advancewidth / 32 * 15(±2)
For Serifs: n advancewidth / 32 * 13(±2)
For example:
If you have a Sans and the /n advancewidth is 400 units, then
400/32*(15-2)= 162 (Smaller deviation)
400/32*15 = 187 (Median)
400/32*(15+2) = 212 (Biggest deviation)
if you have a Serif and the /n advancewidth is 400 units, then
400/32*(13-2)= 137 (Smaller deviation)
400/32*13 = 162 (Median)
400/32*(13+2) = 187 (Biggest deviation)