they're not colourful or textured or on a background or anything. If OCR software can't recognize black type (of a familiar typeface) on white, it must be pretty messed up.

I'm just saying it that making the user do this twice is just plain unusable and annoying.

Anecdotally I've found ReCAPTCHA's CAPTCHA easier to fill in than other CAPTCHA that do strange distortions to obfuscate perfectly rendered text.

It seems the human brain is much more adept at decoding the black and white noise resolved in a crappy scan than the artificial manipulations applied by a computer algorithm.

It also seems the opposite is true for computers.