I like it, sometimes a convention is the only thing that stops us from being creative.

I second four!

I. 3x4 font (which is not quite a tiny font - see '3x3' tag) ; II. 9 dupes(!) / 27 (= ⅓ not quite creative…) ; III. Could be useful as 'fake chars' (substitutive glyphs, for logos etc.) ; IV. Good minimalistic job anyway :-)

@ demonics : never mind. I have at least 20 or 30 years of micro font design or attemps behind me, hence nothing to prove (or to show off for the moment), just advice to the beginning artists. Such a full ASCII 3x3 font exists (at least mine, after dozens of versions and improvements), I cannot lie. Just wait a little, I'll release it to the web, for free I think. (The 3x4 is more appealing and legible though.) And BTW, keep on telling me when I'm wrong in my Cs&Cs, even/esp. if I get a tad arrogant for nothing ! For we were all beginners, and my critics should always be taken as positive, if not an incentive from an old school designer ;-)

PS. Sorry in return, I did not realize immediately you were the designer of this FontStruct ! Yes, it was a bit harsh from me, but I added this comment to all the micro FontStructs I downloaded today (if it can be of some relief - they are all greatly incomplete fonts, unfortunately). Now back to your design : I think an extended dm/lm 2 could take advantage of a bitmap preprocessing in the 1st step of your creation, since this micro font looks like it was automagically rendered after a raster filter was applied to an image of a set of well chosen characters… I hope it makes sense… ImageMagick could do it from a script, which would be more accurate than drawing everything by hand.

PPS. I cannot remember my routine, for it was in the early 90s… but I got a similar microfont (except your neat further vertical tuning) by taking, say, a 8x16 raster font. All its chrs were displayed in one page, then the 'parser' began averaging groups of e.g. 2x2 pixels from this original set. At the end, the micro font was temporarily made of two bitplanes (i.e. 4 grays). Fine tuning manually their curves helped me a lot to get a realistic and uniform 1-bit conversion, e.g. when dealing with your 'c' (by chosing between 2x2 or 3x2, i.e. keeping or not its possible light gray on the right). It's all about computation, as always in esthetics. :-)