Steve Sailer recently reiterated an observation made by Linda Gottfredson that perhaps the surest way to assess a person's IQ from casual conversation without risking a false positive is by asking the person whether or not he likes classical music. If the answer is "yes", his IQ is in the triple-digits. If he doesn't enjoy it, though, his answer is less suggestive.
The brief post's comment thread is pretty heated but largely anecdotal, so let's turn to the GSS for a little quantitative cooling. As fun as the module is (see the Inductivist's famous post on it), the GSS has only directly inquired about musical preferences once, in 1993. The mean wordsum score for that year among whites was 6.24, so it's reasonable to presume those answering 7-10 of the vocab test questions correctly have triple digit IQs, while those answering 5 or fewer correctly are in double-digit territory.
The median score is consistently 6, however, so placing this middling group firmly in either the XX or XXX camp is unsatisfying. Consequently, only the 0-5 and 7-10 cohorts should be considered. Further, foreign-born respondents are excluded to avoid language fluency issues.
The percentage of XXers who say they like classical music "very much" (the strongest affirmative option) is 8.2%, compared to 23.6% of XXXers. Looks like when the person in question is a XXer, you'll get a false positive about one in twelve times.
Among XXXers, 4.2% report "disliking it very much" (the strongest negative option), and only another 9.5% say they merely "dislike" it, the rest either having mixed feelings towards or appreciating it, bringing the total among XXXers who have no use for classical music to 13.7%. By comparison, 41.2% of XXers feel this way. Gottfredson's tactic appears to be one that doesn't return too many false negatives, either.