Intelligence tests are fuzzy, controversial things. They have come up in many of my classes here at the University of Where I Go. I am even taking a class specifically on differences between people, cognitively.
First off, measuring intelligence was not just to pick geniuses out of the crowd (though there have been plenty of "scientists" who tried to do just that). Alfred Binet wrote the first usable test, in order to determine which children should receive special help in school. I think that was pretty awesome and forward-thinking of him.
"IQ" stands for "intelligence quotient," where an individual's test score was divided by his or her age, to produce a ratio of mental age (how well he or she did on the test) to chronological age (how old he or she was). As you can see, this wasn't exactly fair to older people. Say you got a score of 150 on the test, and you are 10 years old. You could say your IQ is 15, whereas someone with the same score who is 50 has an IQ of 3. (IQ tests through the years have had different calculations; I made this one up just to illustrate the point.) Huh? Exactly. Now they are based on bell curves and such and the ageism has been eliminated.
What intelligence really is is a matter of much debate. I think it's pretty safe to say that the various ideas about multiple intelligences, different processes, etc. are much closer to the truth than what an IQ test can show.
But IQ obviously measures something (the skill set for an IQ test is extremely narrow, but is still a skill set) and that something is highly heritable. According to Matt Ridley in his book Genome, if a person takes the test twice, his or her test scores are about 87% correlated (100% being exactly the same, 0% being completely randomly different). It is clear it is the same person taking it. Now, take the scores of identical co-twins raised together and compare. 86%. A statistician would have to concede that he or she couldn't tell if it was the same person or not. In fact, it wouldn't be out of line to declare "Yup, same person." Identical co-twins raised apart have scores that are 76% correlated. To compare, fraternal twins reared together have 55% correlation and biological siblings have 47%. Adopted children living together have 0% correlation.
The higher correlation for fraternal twins probably has something to do with "nature" in the womb, compared to biological siblings, who each have different experiences (their own specific nine months, so to speak) in the womb. The large difference between biological siblings and adoptive siblings clearly points to "nurture" as playing a large role in IQ.