How do you A/B test product names (and/or domain names)?

What criteria do you use to qualify the choices at hand? Some common ones are "be a two-syllable word" and "hard to mistype". But there are other, less obvious ones. For example, I've found sometimes that somewhat controversial and/or non-intuitive names end up being more sticky than something that's easy to get at first but less memorable in the end. Also, what sample size have you found sufficient to go with the winner?